E8400 or Phenom II x3 720???

[Cid Highwind]

Just wanted to say that this debate seriously has a lot of interesting information. I always found it hard to compare the processors of today's consoles with what we have in our PC's, but it really seems there's more to it than them being just general purpose CPUs being put in a closed-spec box.

__________________
PC Specs:
CPU: Intel Q8200 @ 2.8GHz // GPU: Sapphire ATi HD4870 1GB // Mobo: Gigabyte EP35-DS3
RAM: 4GB Kingston HyperX DDR2 1066 // HDD: 1TB Samsung Spinpoint // SPU: Creative X-Fi Xtreme Music
PSU: Hiper Type-R 580W // Monitor: Iiyama B2403WS / 1920*1200

Join the NGEmu Folding @ Home community NOW!

[bkwegoharder]

Originally Posted by Cid Highwind

Just wanted to say that this debate seriously has a lot of interesting information. I always found it hard to compare the processors of today's consoles with what we have in our PC's, but it really seems there's more to it than them being just general purpose CPUs being put in a closed-spec box.

Yeah. It was interesting when I found this stuff out too. But it makes sense when you think about it. If they would have used the strongest pc cpu on the market, the console would get outdated very quickly seeing how the best cpus available during the 360 and ps3s development time were the Athlons and Pentiums. Instead it looked like they sacraficed performance in one area(general purpose processing) so that they can get way more performance in another area(3D). That way the consoles last a lot longer.

It's not like they are specifically weaker or stronger in general. but they are more like tools. The console cpus are tools for one area and the pc cpus are tools for another. The Console cpus are used for gaming, while our cpus aren't specifically made for gaming. On the other hand, we have gpus that a by far more powerful than anything in the consoles so the win edges out to us.

[makotech222]

So youre telling me, that if the staff had say, an extra year, to work solely on the pc version, it would come out with about the same fps as it does now? I Highly doubt that.

Plenty of ports have been struck by early deadlines, take for example SW the force unleashed. That game runs horribly :P

Edit: in addition, plenty of ports, that are done well, play MUCH better on the pc over their console counterparts. Mass effect, Last remnant, RE5, Red faction guerrilla. All of these games look and run better on the PC, so how can the console be "more powerful?"

__________________

Desktop: [Windows 7 x64][Intel E8400@3.8ghz] [DDR2 4gb] [Nvidia EVGA Gtx 260] [Creative X-Fi Soundblaster] [Corsair 550w][Viewsonic 23inch 120hz Display]
Lenovo X200Tablet: [Windows 7 build 7600 x86][L9400 @1.86ghz][Intel GMA 4500MHD][2gb Ram][160GB 7200rpm][LED-lit Touchscreen/pen-abled]

---

Last edited by makotech222; December 22nd, 2009 at 22:51..

[runawayprisoner]

Originally Posted by bkwegoharder

No, you need to read more on this matter. The PPE is that slow with general purpose processing. Yes that is true, but remember general purpose processing doesn't matter much with gaming. The VMX instructions on the ppe and the powerPC G5 are far better than the SSE units on our cpus. The shame is in the benchamarks you posted they specifically said that they weren't testing Altivec performance, and they said they weren't using any of the spes. Of course there are going to be low results with general purpose processing. The only thing I see in any of those benchmarks that even remotely relate to performance in games is the dot product geekbench test. And in that test the ppe turns up 20% faster. But even then, that benchmark doesn't matter at all. When you program correctly for the ps3, you mainly used the ppe to control the spes. So in gaming the ppe shouldn't be used much for the actual game. The 7 spes in the Cell processor by FAR blow the i7 away. The i7 simply isn't made for gaming, none of the x86 processors are. They are made for doing many random things at once.

What about the Xenon in the 360?

And the SPEs themselves have limitations that prevent them from being called real cores. If you say "general purpose" computing does not have anything to do with gaming, I think you really really need to start taking game programming lessons. General purpose has a LOT to do with gaming. :/

Like seriously... it is a CPU. All CPUs are general purpose. Lest it's something else. There's no CPU that is "more suited" to being a gaming CPU by any long stretch. There are just fast and slow CPUs. VMX being more efficient than SSE doesn't mean these processors will get a magical boost from it if the developers do not make use of the extensions, and it's not going to boost performance by over 50% by any really really long stretch.

The PPE isn't just mainly used to control the SPEs... It's the main processing core. Any input has to go through it before going to anywhere else. This is also the reason why developers keep complaining about how hard it is to work on the PS3. Technically, stuffs like Physics calculations or rendering shouldn't be too troublesome because the calculations can all be done on the SPEs and returned in no time, leaving the PPE to only fork out which correct calculations to make. But stuffs like AI are much more complex because they require a lot of branching and forking rather than just pure calculation prowess. And that's where the PPE falls short. It just can't keep up with that.

that is not true. again, the benchmarks you posted don't test the vector extensions of the PowerPC or the PPE. The second benchmark you posted compared the PowerPC to the pentium and the Athlon. But like I said before the comparison of the PPE to the PowerPC 1.6GHz was with mainly general purpose processing benchmarks and they didn't even use the vector instruction sets of the ppe or the PowerPC 1.6GHz so the benchmarks don't accurately show gaming performance.

The VMX instruction set is mainly used for accelerating calculations of floating point numbers, which are not very necessary unless you're calculating physics or 3D graphics (or decoding some video formats). Gaming performance will not get a magical boost from it... because sincerely, it's just another set of instructions. Performance is still dependent upon the CPU. The PPE in the Cell BE doesn't even seem to have VMX (that's probably why they didn't use it for the test). It's the SPEs that have VMX.

This is like saying PCSX2 takes a 100% performance hit for running on SSE2 and not SSSE3.

...it just doesn't work like that.

The VMX instructions on the PPE are much better than the SSE units on the Core 2 Duo. And that is what would actually matter with gaming. Even though the SSE units on our cpus are not used fully because of compatability with cpus from different companies. On the 360, the units would be used to their fullest if you want to get the best performance. Plus the Xenon has two VMX 128 units per core. The VMX 128 is a lot faster than the regular VMX instructions on the PPE(It's one of the "little tweaks" that the Xenon got over the ppe.) To be exact, the Xbox 360 has two VMX 128 units per core. So that means 256 registers on each core vs just 8 registers per core with SSE. So for the whole Xenon, there are 768 registers vs 16 in a Core 2 Duo. The VMX 128 instructions were greatly enhanced over VMX specifically to improve 3d and physics performance.

Uhh... you need to really, really, REALLY check your sources, too.

Just like being there: Papers from the Fall Processor Forum 2005: Application-customized CPU design

IBM says there's only one VMX/FPU unit, even though it does have access to 128 registers that are compliant with VMX128.

Weird, though. Wikipedia links to that page and they still say it's 2 FPUs per core... Someone needs to fix that.

Edit: Oh... I see what happened. People look at the diagram and thought that there were two FPUs. There's only one for each core. It can work as either a VMX or FPU unit. Whichever way, it's still an FPU. Now that... ehh... means performance is again very dependent upon the FPU.

So lets take your Core 2 duo @ 2GHz and measure up the maximum theoretical 3D performance it gets with SSE. We know it is 4flops per clock cycle. So @ 2ghz it would be 8GFlops per core. So all in all 16Gflops. Vs the Xenon using the regular VMX instructions. 3.2GHz * 8flops per cycle=25.6 per core. 76.8GFLOPs for the entire cpu using just the vmx instructions. So this shows that the performance in vector operations is by far better on a 3 core PPE. Hell it's better than the Core 2 quad and the i7. And all of that doesn't even take into account that the 360 has the VMX 128 instructions vs the regular VMX instructions.

Do you use the CPU to render 3D graphics in this case? If not... then this data is completely irrelevant and won't be any helpful to gaming at all. And again... just marketing BS. Use real-life benchmark data.

If VMX was so good, you'd have seen consoles outdo PCs in terms of physics ages ago. But they don't. It's simply because maximum theoretical performance is never reached. Especially in a multicore design. If it's like you said then a Core 2 Quad has to be twice as fast as a Core 2 Duo. Is it?

For 3D performance, the 360 processor would easily crush a Core 2 duo @ 2GHz.

Yes, yes, 3D performance. Again... do you, like DO YOU, render on the CPU, or on the GPU?

...I think we can reach an agreement there.

The Dashboard uses an entire core? Please show me proof. Because I originally thought that it was just the ps3 that uses a whole spe for the OS. I don't see anything over the internet that says it reserves an entire core for the dashboard.

It doesn't reserve an entire core. I didn't say it reserves an entire core, but in-game Dashboard means that it's still running in the background and taking up some resources. Unless you're saying the Dashboard is running for free.

That is far from the truth. Like I said before I was able to use just one core of my i7 and run the game with 360 settings at the same frame rate.

Say that to the hundreds of users who are struggling to do the same thing on different system configurations. And how did you even test that?

You could even look on youtube for some videos of a 2GHz Core 2 duo running gta 4 at well above 360 settings .

Do they run at a smooth 30fps in spots where the 360 run at said framerate? Or is it all hearsay?

No. you do YOUR research. You went and took benchmarks for general purpose processing and tried to use them in an arguement about gaming performance. And gaming performance has barely anything to do with general purpose processing, and it has everything to do with SIMD/vector processing.

Ah... gee... Again, ALL CPUS ARE GENERAL PURPOSE. There's no such thing as a gaming CPU. If it's leaning towards computing a specific type of data then it's not a CPU, it's just a processor. Vector processing has very little to do with gaming except for physics calculations and maybe some multimedia processing. But AI and the rest of the other mumble jumbles (occlusion culling, for instance) still run on your raw, plain CPU. That's why developers keep complaining about the Cell BE.

If it was truly as "powerful" as you are trying to make it out to be then nobody would complain about how hard it is to program for the Cell BE because then they don't even need to touch the SPEs for extra power.

Or do you have an actual article or source that says that general purpose computing has absolutely nothing to do with gaming?

Plus, the benchmarks are not even optimized for in order processors so the processor probably took a very huge performance penalty for that. Any good game developer for the ps3 and 360 would optimize their code for in order processing, and that would greatly minimize the difference between the performance of an in order processor.

Do you really know the difference between an in-order processor and an out-of-order processor...?

Out-of-order execution - Wikipedia, the free encyclopedia

This would be costly if the frequency at which the processor is running is very low, but as the frequency gets higher, the latency can be neglected and the penalty is not as high. In-order processors also get a huge performance boost by increasing the frequency even if just by a little.

To remedy the situation, a virtual hardware thread would work about the same as a substitute for out-of-order execution. If one thread is stalled because the operands or parameters of an instructions haven't arrived yet, the other thread will get executed.

Actually the guy who ran the benchmarks went and admitted that they weren't using the full capability of the processor. Hell even the guys commenting on the article even noted that.

Geek Patrol | PlayStation 3 Performance

You took the benchmark results out of context. The benchmarks barely make use of even the ppe. Let alone the spes.

Full use means they have to use the SPEs as well. They are trying to benchmark the whole Cell BE, not just the PPE. In that context, yes, their result is not satisfactory because the whole Cell BE doesn't perform on the same level as a G5. But the PPE definitely performs on the same level as a G5 and even with the help of a VMX/FPU unit, it doesn't get significantly faster. 100% faster is at most as it can do, but then it'll just be as fast as your average Core 2 at 2GHz.

Again, that is only with out of order general purpose code. the benchmarks have nothing whatsoever to do with 3d performance and they don't even test the full capabilities of the cpu. Even the author admitted that.

Read the above please... I really can't believe I have to pull articles for you to read about something you could have researched before you make statements.

Tweaked SOMEWHAT? WTH? They added 768 freaking registers to the cpu. That is not tweaked somewhat. Well, actually it is tweaked somewhat since they did just add 2 enhanced VMX units per core, while leaving the rest of the cpu(except for the cache and probably some other little details.) The performance improvement would be much greater than adding a measily 400MHz. If it weren't the case then wouldn't microsoft and IBM simply up the clockrate slightly instead of adding 2 full blown vector units? Adding two vector units would increase the die size, add a lot of complexity to the design, cost them more money etc.

Read above again, please. Direct from IBM. Or am I just pulling stuffs out of my A?

---

Last edited by runawayprisoner; December 23rd, 2009 at 01:13..

[Phil]

Interesting discussion

Even though they are apples and oranges lol.

A Cell is good at rendering, not specifically "gaming". LOL!

Just for your information, consoles have been getting closer and closer to PC's.....a garden variety NES has more proprietary parts then your entry level 360.

__________________

Don't argue with an idiot, because he will drag you to his own level and beat you with experience.

[bkwegoharder]

Originally Posted by runawayprisoner

What about the Xenon in the 360?

Like I said before, the Xenon is stronger than the ppe because of the vmx units. Since there are 3 of them and not just one, the Xenon can hold it's own.

And the SPEs themselves have limitations that prevent them from being called real cores. If you say "general purpose" computing does not have anything to do with gaming, I think you really really need to start taking game programming lessons. General purpose has a LOT to do with gaming. :/

General purpose code does have a lot to do with gaming. BUT general purpose code takes a a very small amount of cpu cycles to execute compared to other things like physics and any thing relating to graphics. AI is the hardest thing for the cpu to do when it comes to general purpose code, and even then AI is very easy to do on even a weak processor.

Like seriously... it is a CPU. All CPUs are general purpose. Lest it's something else. There's no CPU that is "more suited" to being a gaming CPU by any long stretch.

You obviously don't now the definition of a cpu then. General purpose processors are just one type of cpu. The SPEs are another type. They are vector processors. Here is wikipedia if you don't believe me.

Vector processor - Wikipedia, the free encyclopedia


A vector processor, or array processor, is a CPU design wherein the instruction set includes operations that can perform mathematical operations on multiple data elements simultaneously.

Our cpus from Intel or AMD and the Xenon are on the otherhand Superscalar cpus.

Superscalar - Wikipedia, the free encyclopedia

. VMX being more efficient than SSE doesn't mean these processors will get a magical boost from it if the developers do not make use of the extensions

The thing is why would they not make use of the VMX extensions. If you are trying to make full use of the system, then you will have to use the VMX instructions. If you don't then you aren't using the full power of the cpu.

and it's not going to boost performance by over 50% by any really really long stretch.

Any proof for that? I just proved the exact opposite in my last post. Depending on what the cpu is doing, a single vmx unit can do things many times faster.

The PPE isn't just mainly used to control the SPEs... It's the main processing core. Any input has to go through it before going to anywhere else. This is also the reason why developers keep complaining about how hard it is to work on the PS3. Technically, stuffs like Physics calculations or rendering shouldn't be too troublesome because the calculations can all be done on the SPEs and returned in no time, leaving the PPE to only fork out which correct calculations to make.

But stuffs like AI are much more complex because they require a lot of branching and forking rather than just pure calculation prowess. And that's where the PPE falls short. It just can't keep up with that.

Yeah, the thing is AI is already ridiculously easy to execute so it's no problem for the ps3. The 360 cpu is 3 times better than the ps3 withsince it has 3 times the general purpose processing power, but that doesn't matter since AI, Game logic, and other things that need general purpose computing like player input(the controllers basically) simply don't need a strong cpu to execute. They do usually make up most of the code for the game, but they are done very quickly by any recent cpu. In the end the ppe can easily handle any general purpose code that a game would need to execute.

The VMX instruction set is mainly used for accelerating calculations of floating point numbers, which are not very necessary unless you're calculating physics or 3D graphics (or decoding some video formats). Gaming performance will not get a magical boost from it... because sincerely, it's just another set of instructions. Performance is still dependent upon the CPU.

So you are saying that the VMX instructions don't do anything for gaming, even though you just said that they are good for 3D graphics and Physics? Don't video games have 3D graphics and physics?

The PPE in the Cell BE doesn't even seem to have VMX (that's probably why they didn't use it for the test). It's the SPEs that have VMX.

Do your research. It does have the VMX instructions. It doesn't however have the VMX 128 instructions.

The SPEs don't have VMX units. Hell they don't even need VMX units because they are already SIMD processors. Why would you add a SIMD processing unit to a SIMD processor?

This is like saying PCSX2 takes a 100% performance hit for running on SSE2 and not SSSE3.

...it just doesn't work like that.

First off, SSE2 to SSSE3 is not a major jump. It is barely a hop. You can't compare the difference between VMX and the SSE instructions to the difference between SSE2 and SSSE3.

Also, to use turn your argument right back at you. Look at the performance of PCSX2 when running on a processor with SSE2 vs a processor with just SSE. PCSX2 barely works without SSE2.

Uhh... you need to really, really, REALLY check your sources, too.

Just like being there: Papers from the Fall Processor Forum 2005: Application-customized CPU design

IBM says there's only one VMX/FPU unit, even though it does have access to 128 registers that are compliant with VMX128.

Weird, though. Wikipedia links to that page and they still say it's 2 FPUs per core... Someone needs to fix that.

Edit: Oh... I see what happened. People look at the diagram and thought that there were two FPUs.

No the Wikipedia page is right. I misunderstood. There aren't 2 vmx units per core. I thought that there were though. The diagram clearly shows 2*128*128bit registers in the VRF(the register files) part of the VMX 128 unit. So basically 2 sets of 128 128bit registers. I assumed it meant VMX units. I assumed it meant 2 vmx units when wikipedia showed 2*128*128bit. My bad. My point still stands though. A Single vmx unit easily out performs the SSE units in a Core 2 duo @ 2.0GHz.

There's only one for each core. It can work as either a VMX or FPU unit. Whichever way, it's still an FPU. Now that... ehh... means performance is again very dependent upon the FPU.

The VMX unit is a combined SIMD(vector) and FPU unit. You can't just simplfy it to just an FPU. That would be like calling the Cell processor a general purpose processor.

Do you use the CPU to render 3D graphics in this case? If not... then this data is completely irrelevant and won't be any helpful to gaming at all. And again... just marketing BS. Use real-life benchmark data.

Yes you do use the cpu to render 3D graphis. That is the entire point of adding the VMX128 unit. Just like how with the Cell processor you are supposed to use the spes. With the 360, to get the best performance out of the entire console. You are supposed to use the cpu in conjunction with the video card.

If VMX was so good, you'd have seen consoles outdo PCs in terms of physics ages ago. But they don't. It's simply because maximum theoretical performance is never reached. Especially in a multicore design. If it's like you said then a Core 2 Quad has to be twice as fast as a Core 2 Duo. Is it?

A Core 2 quad is 2 times better than the Core 2 duo. Developers are simply just starting to even multithread their games. Just give it time. I get slightly more than double the performance when going from a single core to dual core in GTA 4.

The thing you said about theoretical performance is true. But look at that apple link I posted describing the vmx instructions. They were able to get 95.5% of the theoretical performance from VMX. Also, the raw theoretical performance of the Xenon is many times higher than the Core 2 Duo @ 2GHz, so even if the Core 2 Duo @ 2.0GHz(I'll just say e4400 from now on since it is clocked at 2.0GHz.) can get the full 100% of the theoretical performance out of it, and the 360 cpu xenon can only get 25%, the Xenon would still be many times better.

As for why no console games have surpassed the pc in terms of physics. I don't know why. It could be because the cpus are already being used for 3D graphics in conjunction with the gpus, while 3d is mainly rendered on the gpu on our pcs. So the console cpus have to handle more.

Or it could be that most developers are mainly using middleware like Havok, PhysX and Bullet on the consoles and not many are pushing the envolop like how Crytek did with Crysis. Who knows?

Crysis however was shown on the consoles and the physics effects shown were very very good and all real time. And certain displays of physics looked even better than Crysis 1.

Yes, yes, 3D performance. Again... do you, like DO YOU, render on the CPU, or on the GPU? ...I think we can reach an agreement there.

On the Xbox 360 and ps3, you render on BOTH. That is the huge difference between our pcs and the consoles. The consoles have ****ty gpus so you have to use both to get the best performance. Our gpus are already very very fast so we don't need to do 3D graphics on the cpu.

It doesn't reserve an entire core. I didn't say it reserves an entire core, but in-game Dashboard means that it's still running in the background and taking up some resources. Unless you're saying the Dashboard is running for free.

Okay. You dind't say it reserves a core. You said "(note that Xenon also has to run the Dashboard in the background with one core)"

Honestly I don't see why you even brought it up. The Dashboard is very light and has a very small impact of performance, especially compared to what Windows.

Say that to the hundreds of users who are struggling to do the same thing on different system configurations. And how did you even test that?

I will say that to them. If you can't run the game at 360 settings, your pc is either very weak, or it is something wrong with it. I tested it by going into my bios and shutting off all cores except for one, while leaving hyperthreading on.

Do they run at a smooth 30fps in spots where the 360 run at said framerate? Or is it all hearsay?

I know that a Core 2 Duo @ 2GHz would have to be around the performance of a single core i7 with hyperthreading @ 2.66GHz. I also get hearsay about it.

Ah... gee... Again, ALL CPUS ARE GENERAL PURPOSE. There's no such thing as a gaming CPU. If it's leaning towards computing a specific type of data then it's not a CPU, it's just a processor.

I already addressed this.

Vector processing has very little to do with gaming except for physics calculations and maybe some multimedia processing.

And 3D graphics. Hell the guys who made The sabeteour used the spes to do anti aliasing. The guys who made killzone 2 used the spes to do deffered rendering. etc. Vecor processors are also much faster at video encoding. A Core i7 965 lost to the Cell when encoding a 1080p video. Vector processors are better for almost any form of Multimedia. Don't try to downplay it by saying. "maybe some multimedia processing"

But AI and the rest of the other mumble jumbles (occlusion culling, for instance) still run on your raw, plain CPU. That's why developers keep complaining about the Cell BE.

AI is very easy to run. I haven't seen any game developer say that they were having an issue with their game running ai on the ps3.I have heard some early speculation about it before the ps3 released. But I've heard nothing about it in active developement. actually, Killzone 2 has very great AI. Yet that is a PS3 game. Look at Heavy Rain and how realistic the people act. The AI arguement is overhyped. As for occlusion culling. I haven't looked into it and I'm ignorant on that specific subject so I'm not really going to reply on that now.

I can say however that the main reason I'm see developers complaning about the Cell is because of how complex the design is and how hard it is to program for compared to the ps3.

If it was truly as "powerful" as you are trying to make it out to be then nobody would complain about how hard it is to program for the Cell BE because then they don't even need to touch the SPEs for extra power.

A single ppe is weak. There is just one. and compared to even one core in the Xenon it is not very strong. Let alone 3 so if you don't make use of the spes, you'll get lack luster performance. Okay so it beats a Core 2 Duo at 2.0GHz in 3d. For one beating a Core 2 duo @ a low clock rate like 2.0GHz is nothing to get excited about. It also doesn't matter because on the pc you don't use the cpu to do 3d anyway so. Compared to the 360 tri core cpu the ppe by it self is very weak.

Or do you have an actual article or source that says that general purpose computing has absolutely nothing to do with gaming?

Actually I do have something about that.

PS3 Hardware: Explained - *Revision 1* - PS3 Forums

I also remember seeing some debug information about killzone 2 that developers had shown and it showed that ai did use a very small amount a cpu time, it also showed the specifics of how much percentage each aspect of what they were doing was using up of each part of the ps3 including the gpu, the ppe, the spes, the ram etc.

I'm looking for it now I'll post it later.

I'll admit, I did over exaggerate when I said, it has "nothing" to do with gaming. But the vast majority of what really takes up resources is 3D graphics.

Do you really know the difference between an in-order processor and an out-of-order processor...?

Out-of-order execution - Wikipedia, the free encyclopedia

This would be costly if the frequency at which the processor is running is very low, but as the frequency gets higher, the latency can be neglected and the penalty is not as high. In-order processors also get a huge performance boost by increasing the frequency even if just by a little.

To remedy the situation, a virtual hardware thread would work about the same as a substitute for out-of-order execution. If one thread is stalled because the operands or parameters of an instructions haven't arrived yet, the other thread will get executed.

Yes I know that. But if the cpu is running code optimized for out of order execution, then the performance will still be greatly impacted even with the multithreading and high clock rate.. There are ways to cut down on the impact, but it doesn't change the fact that on the software side, the code should be optimized for in order execution. Once the code is optimized for in order execution, the performance difference between a processor using out of order execution can be reduced significantly.

Full use means they have to use the SPEs as well. They are trying to benchmark the whole Cell BE, not just the PPE. In that context, yes, their result is not satisfactory because the whole Cell BE doesn't perform on the same level as a G5. But the PPE definitely performs on the same level as a G5 and even with the help of a VMX/FPU unit, it doesn't get significantly faster. 100% faster is at most as it can do, but then it'll just be as fast as your average Core 2 at 2GHz.

The G5s have the VMX extensions as well. Because of them, a Dual 2.5GHz G5 was about to equal a Core 2 Quad q6600 in multimedia. According to the link you posted a dual core g5 @ 2.5GHz equals a Core Duo(not even a core 2 duo but a regular core duo) @ 1.83GHz. That is way way bigger than a 100% performance boost. Looking at what intel says about a Core Duo @ 1.83GHz, the maximum theoretical performance is 5.49Gflops.

Look here

Processors - Intel® microprocessor export compliance metrics

The maximum theoretical performance of the q6600 is 38.4Gflops. Remember I said theoretical performance.

Hardware - SIMD Executive Summary

38.2Gflops for the g5 in a REAL benchmark. That isn't even theoretical performance.

Don't put random limitations on vmx units. Depending on what you are doing the boost in performance can be either very very large or no boost at all.

Read the above please... I really can't believe I have to pull articles for you to read about something you could have researched before you make statements.

Why don't you read the above please. I can't believe I have to pull articles for you to read about something you could have researched before you made statements, even AFTER I already have shown that the articles you chose are completely irrelevant.

Read above again, please. Direct from IBM. Or am I just pulling stuffs out of my A?

I guess you are because if you look clearly at the diagram at the vrf(The vector Register Files) it says 2*128*128bit. So yes. there are 256 registers per vmx unit and 3 vmx units per core. 768 registers. I was mistaken on thinking that there were two vmx units per core. There are not, but there are 2 sets of 128 128bit registers

Originally Posted by makotech222

So youre telling me, that if the staff had say, an extra year, to work solely on the pc version, it would come out with about the same fps as it does now? I Highly doubt that.

What? When did I say that? How does that even relate to what I was saying?

Edit: in addition, plenty of ports, that are done well, play MUCH better on the pc over their console counterparts. Mass effect, Last remnant, RE5, Red faction guerrilla. All of these games look and run better on the PC, so how can the console be "more powerful?"

I never said the consoles were more powerful. I did say that the cpus were more powerful than our desktop, for CERTAIN jobs. All in all a modern pc will easily crush the consoles because the gpus in the consoles are weak as hell.

---

Last edited by bkwegoharder; December 23rd, 2009 at 08:12.. Reason: Automerged Doublepost

[makotech222]

youre saying gta4 is optimized, so if they spent another year working on it, it wouldnt make a difference in the results?

__________________

Desktop: [Windows 7 x64][Intel E8400@3.8ghz] [DDR2 4gb] [Nvidia EVGA Gtx 260] [Creative X-Fi Soundblaster] [Corsair 550w][Viewsonic 23inch 120hz Display]
Lenovo X200Tablet: [Windows 7 build 7600 x86][L9400 @1.86ghz][Intel GMA 4500MHD][2gb Ram][160GB 7200rpm][LED-lit Touchscreen/pen-abled]

[runawayprisoner]

Originally Posted by bkwegoharder

Like I said before, the Xenon is stronger than the ppe because of the vmx units. Since there are 3 of them and not just one, the Xenon can hold it's own.

Uhh... proof? Again, the VMX units are for calculating floating point numbers only. Most games would depend on integer performance rather than floating point (unless it's physics calculations) so again, they are not going to be miraculous boosters. Even then, there are no data on VMX128 versus VMX performance so there's no telling if one core of Xenon is faster than the PPE or not.

General purpose code does have a lot to do with gaming.

You just said it was irrelevant to gaming.

BUT general purpose code takes a a very small amount of cpu cycles to execute compared to other things like physics and any thing relating to graphics.

...If you say that to a developer, he'll smack you upside down the head. Really. General purpose codes take just as many cycles as physics and anything related to graphics to execute. Actually, sometimes even more cycles. Things like iterating through an array or looking up a specific symbol within an array are very cycle-intensive. It's mostly just sorting math when it comes to graphics related things (this is only to take off a bit of burden for the GPU).

Physics calculations shouldn't take that many cycles, either, unless the CPU lacks a proper FPU. In that case, you can still simulate floating point calculations through integers... and that's what makes these things cycle intensive (since you have to convert integers back and forth). But if the developers choose to do inexact physics (hit tests, for instance), or math with very large integers (in this case, a conversion to floating point number will just be redundant), then the number of cycles the CPU has to go through is reduced significantly. This will also eliminate the need to use single-precision float because it'll just be a tiny bit more accurate.

AI is the hardest thing for the cpu to do when it comes to general purpose code, and even then AI is very easy to do on even a weak processor.

Uhh... check your facts again, please. AI is not very easy to do on a weak processor. If you're talking about very simple AI (like in the asteroid types of games), then you're right. But if you're talking about advanced AI, then those are extremely CPU-heavy. And depending on the complexity of the game engine, the AI has to calculate a lot more. This shows the most in FPS games where the AI has to calculate the path that it takes as well as the trajectory of its shots and everything else related to that. About 10 of those things visible on screen and the burden increases ten-fold. If AI calculation was so easy, then Crysis and Far Cry 2 wouldn't get a boost from a faster CPU even though Physics were turned off.

You obviously don't now the definition of a cpu then. General purpose processors are just one type of cpu. The SPEs are another type. They are vector processors. Here is wikipedia if you don't believe me.

Vector processor - Wikipedia, the free encyclopedia


A vector processor, or array processor, is a CPU design wherein the instruction set includes operations that can perform mathematical operations on multiple data elements simultaneously.

Our cpus from Intel or AMD and the Xenon are on the otherhand Superscalar cpus.

Superscalar - Wikipedia, the free encyclopedia

That's great... but... the SPEs are still not full-fledged CPUs. They don't act like that. The most common mistake anyone can make is by thinking that they are CPUs. They act more like co-processors. If they were CPUs, then you could just bypass the PPE and get to the SPEs directly. Unless I'm missing something.

Also the same Wikipedia said earlier that the Xenon had 2 vector units per core. I would take Wikipedia with a 50/50 mindset and check its sources before I make statements. Even the main Cell BE article on Wikipedia does not mention the SPEs as being vector processors.

The thing is why would they not make use of the VMX extensions. If you are trying to make full use of the system, then you will have to use the VMX instructions. If you don't then you aren't using the full power of the cpu.

What will they make use of the VMX extensions for...? Even in games, they are not always necessary.

Any proof for that? I just proved the exact opposite in my last post. Depending on what the cpu is doing, a single vmx unit can do things many times faster.

The GFLOPS of the PS3 performing single-precision floating point calculations is up to 200. Yeah, that's great. But did you know that if you want to do precise physics calculations, you use double-precision floating point calculations? And when you do so, even the mighty Cell BE can only reach around 20GFLOPS or so. Here's a wikipedia quote since you like to check that place so much:

Cell (microprocessor) - Wikipedia, the free encyclopedia

And here's a quote, again, from Wikipedia:

FLOPS - Wikipedia, the free encyclopedia

Your fastest Core i7 can reach 70GFLOPS in double precision. It that regard, it's actually 3 times faster than Cell BE.

The PPE in the Cell BE is rated for 6.4GFLOPS when it comes to double precision. So that's about 19.2GFLOPS with the Xenon. That is actually about as many GFLOPS as a 2.4GHz Core 2 Duo can dish out. Not mentioning a Core 2 Quad, of course. And not to mention the Core 2 Duo will smoke Xenon in general purpose tasks any day.

Yeah, the thing is AI is already ridiculously easy to execute so it's no problem for the ps3. The 360 cpu is 3 times better than the ps3 withsince it has 3 times the general purpose processing power, but that doesn't matter since AI, Game logic, and other things that need general purpose computing like player input(the controllers basically) simply don't need a strong cpu to execute. They do usually make up most of the code for the game, but they are done very quickly by any recent cpu. In the end the ppe can easily handle any general purpose code that a game would need to execute.

Actually, the ONE thing that drove Valve away from further releasing Source games on the PS3 was because the CPU could not keep up with the engine. Games like Left 4 Dead (up to 30 visible zombie AIs at the same time) would be almost impossible to do, and even on the 360, such games can only do 30fps at most.

Please check your facts.

So you are saying that the VMX instructions don't do anything for gaming, even though you just said that they are good for 3D graphics and Physics? Don't video games have 3D graphics and physics?

I said that they don't do much. At least not to the point where you are trying to make them out to be. Physics calculations require very precise calculations. Most of the time with double precision float. That's why you don't see the likes of PhysX or Havoc on consoles as you do on PC. Xenon or Cell BE, whatever, are really really horrible at double-precision floating point operations. The situation is somewhat better with Cell BE and we can at least get performance close to PC, but it's almost a lost cause with 360.

As for 3D graphics, it's only good if the CPU is what's rendering the scene. Unfortunately, the rendering is done on the GPU so the rest of the instruction extensions just seem really redundant.

Do your research. It does have the VMX instructions. It doesn't however have the VMX 128 instructions.

The SPEs don't have VMX units. Hell they don't even need VMX units because they are already SIMD processors. Why would you add a SIMD processing unit to a SIMD processor?

Well, I'm wrong there. Read it backwards. But that still doesn't help.

First off, SSE2 to SSSE3 is not a major jump. It is barely a hop. You can't compare the difference between VMX and the SSE instructions to the difference between SSE2 and SSSE3.

Yeah, yeah... BUT, oh my God... what do you think VMX is used for? No, wait, actually, quote a game developer who says that the VMX instruction set in the Xenon works miracles.

Also, to use turn your argument right back at you. Look at the performance of PCSX2 when running on a processor with SSE2 vs a processor with just SSE. PCSX2 barely works without SSE2.

No the Wikipedia page is right. I misunderstood. There aren't 2 vmx units per core. I thought that there were though. The diagram clearly shows 2*128*128bit registers in the VRF(the register files) part of the VMX 128 unit. So basically 2 sets of 128 128bit registers. I assumed it meant VMX units. I assumed it meant 2 vmx units when wikipedia showed 2*128*128bit. My bad. My point still stands though. A Single vmx unit easily out performs the SSE units in a Core 2 duo @ 2.0GHz.



The VMX unit is a combined SIMD(vector) and FPU unit. You can't just simplfy it to just an FPU. That would be like calling the Cell processor a general purpose processor.

Yes you do use the cpu to render 3D graphis. That is the entire point of adding the VMX128 unit. Just like how with the Cell processor you are supposed to use the spes. With the 360, to get the best performance out of the entire console. You are supposed to use the cpu in conjunction with the video card.

...Uhmmmm... wow. Do you know that if you use the CPU to render 3D graphics then you have to use the CPU to complete at least that rendering task before sending it to the GPU? That's dead slow. No, wait, that's not just dead-slow, that's crawling. You need to link up 2 or more PS3s to be able to get around 30fps rendering complete scenes all by themselves at game resolutions. How "fast" do you think the Xenon CPU is to be able to render 3D graphics along with the GPU?

A Core 2 quad is 2 times better than the Core 2 duo. Developers are simply just starting to even multithread their games. Just give it time. I get slightly more than double the performance when going from a single core to dual core in GTA 4.

Wow... I'll just let other people comment on this.

The thing you said about theoretical performance is true. But look at that apple link I posted describing the vmx instructions. They were able to get 95.5% of the theoretical performance from VMX. Also, the raw theoretical performance of the Xenon is many times higher than the Core 2 Duo @ 2GHz, so even if the Core 2 Duo @ 2.0GHz(I'll just say e4400 from now on since it is clocked at 2.0GHz.) can get the full 100% of the theoretical performance out of it, and the 360 cpu xenon can only get 25%, the Xenon would still be many times better.

Read above, please. Single-precision math, it may be faster. But double-precision math... erg. Again, these things are not miracle workers. Granted, Core 2 is not that much faster if they have to do single-precision math, but double-precision math is more used in physics simulations because it's more accurate that way, and both Cell BE and Xenon take humongous (up to 90% performance hit) when calculating double-precision. That's why developers keep saying PS3 and 360 perform similarly and 360 has a better GPU (Cell BE and Xenon both have around 20GFLOPS of performance when it comes to double-precision) even though we keep seeing Cell BE being 3 times faster than Xenon.

[QUOTE}As for why no console games have surpassed the pc in terms of physics. I don't know why. It could be because the cpus are already being used for 3D graphics in conjunction with the gpus, while 3d is mainly rendered on the gpu on our pcs. So the console cpus have to handle more. [/quote]

Read above, please.

Okay. You dind't say it reserves a core. You said "(note that Xenon also has to run the Dashboard in the background with one core)"

Honestly I don't see why you even brought it up. The Dashboard is very light and has a very small impact of performance, especially compared to what Windows.

Eh... you forgot the fact that Xenon is based on in-order processing. The Dashboard might be light, but it's not so light that the processor doesn't take some hits from it.

I will say that to them. If you can't run the game at 360 settings, your pc is either very weak, or it is something wrong with it. I tested it by going into my bios and shutting off all cores except for one, while leaving hyperthreading on.

Turn even HyperThreading off then.

I know that a Core 2 Duo @ 2GHz would have to be around the performance of a single core i7 with hyperthreading @ 2.66GHz. I also get hearsay about it.

No, a Core 2 Duo @ 2GHz is equal to a Core i7 without HyperThreading running on 2 cores at 1.93GHz.

And 3D graphics. Hell the guys who made The sabeteour used the spes to do anti aliasing. The guys who made killzone 2 used the spes to do deffered rendering. etc. Vecor processors are also much faster at video encoding. A Core i7 965 lost to the Cell when encoding a 1080p video. Vector processors are better for almost any form of Multimedia. Don't try to downplay it by saying. "maybe some multimedia processing"

So you do 1080p encoding while you play a game?

That's point one. On PC, you can just throw everything onto the GPU and be done with it (what CryEngine 2 does). That's the mindset.

AI is very easy to run. I haven't seen any game developer say that they were having an issue with their game running ai on the ps3.I have heard some early speculation about it before the ps3 released. But I've heard nothing about it in active developement. actually, Killzone 2 has very great AI. Yet that is a PS3 game. Look at Heavy Rain and how realistic the people act. The AI arguement is overhyped. As for occlusion culling. I haven't looked into it and I'm ignorant on that specific subject so I'm not really going to reply on that now.

Uh... they don't just have AI problems with the Cell BE. They have ALL sorts of problems with the Cell BE. Technically, they have problems processing everything, not just AI. AI plays a part, though.

If not then Valve wouldn't complain and quit the PS3.

I can say however that the main reason I'm see developers complaning about the Cell is because of how complex the design is and how hard it is to program for compared to the ps3.

The design is not more complex than a Core 2 Quad or anything like that. The problem is that the CPU is too slow at some tasks that would otherwise run very fast on some other systems. But they can't just say that the Cell BE because then they'd get laughed at, so they just chalk it off as complex design.

Actually I do have something about that.

PS3 Hardware: Explained - *Revision 1* - PS3 Forums

I also remember seeing some debug information about killzone 2 that developers had shown and it showed that ai did use a very small amount a cpu time, it also showed the specifics of how much percentage each aspect of what they were doing was using up of each part of the ps3 including the gpu, the ppe, the spes, the ram etc.

I'm looking for it now I'll post it later.

I'll admit, I did over exaggerate when I said, it has "nothing" to do with gaming. But the vast majority of what really takes up resources is 3D graphics.

Yes I know that. But if the cpu is running code optimized for out of order execution, then the performance will still be greatly impacted even with the multithreading and high clock rate.. There are ways to cut down on the impact, but it doesn't change the fact that on the software side, the code should be optimized for in order execution. Once the code is optimized for in order execution, the performance difference between a processor using out of order execution can be reduced significantly.

Uhh... there's no code "optimized" for out-of-order or in-order execution. They are just different execution style. Obviously... very compact codes will run fast in both, while very heavy codes will run slower on in-order.

It's still up to developers. In some cases, you just can't avoid heavy codes.

Don't put random limitations on vmx units. Depending on what you are doing the boost in performance can be either very very large or no boost at all.

It's not what I'm doing. It's what developers do. And if they've done as you've said that they should do, then you'd see consoles outperforming PCs by miles when it comes to physics. They don't. And I put the reason above.

I guess you are because if you look clearly at the diagram at the vrf(The vector Register Files) it says 2*128*128bit. So yes. there are 256 registers per vmx unit and 3 vmx units per core. 768 registers. I was mistaken on thinking that there were two vmx units per core. There are not, but there are 2 sets of 128 128bit registers

There is only one set. Because VMX 128 can only work with 128 registers at the same time. What the diagram meant to say was 2 threads, both able to access the common 128 registers at 128-bit length for each register.

I never said the consoles were more powerful. I did say that the cpus were more powerful than our desktop, for CERTAIN jobs. All in all a modern pc will easily crush the consoles because the gpus in the consoles are weak as hell.

Not just for certain jobs. For no jobs at all. It's all just numbers on papers. Granted, Cell BE will do better than "some" desktops at Folding@Home, but that's it. Nothing else console has done, or will ever do, can outperform PCs due to:
- The CPUs are horrible at double-precision maths, which are most often used for physics simulations. They're not that horrible at single-precision, but theoretical performance =/= practical performance.
- VMX instructions are cool and all, but if multicore support from developer sucks then performance also suffers. The past years have shown that for us. Even with graphics chips with thousands of shader processors, efficiency is not at 100%. If they were then HD4870 would smoke GTX 285 by miles.
- 3D rendering is not done on the CPU... at least not on Xenon because the CPU is not fast enough for that. On the PS3, the SPEs only do the preparing work. The rendering still has to be done on the GPU. That's how it's done on PC, too, though.
- A Core i7 will outperform Cell BE or Xenon at double-precision by many folds. That's because double-precision counts more than single-precision.
- Just a fun note: Flash also runs horrible on the PS3 even though it's touted as this very fast CPU. Simply because all Flash maths involve double-precision.

[Phil]

Ive never seen more WOT on any one place on this forum

__________________

Don't argue with an idiot, because he will drag you to his own level and beat you with experience.

[Yogi]

Woohooo....

My brain is exploding & my eyes are bleeding!! Never in my life i'm kneeling before a wall of TEXT!!

[bkwegoharder]

Originally Posted by runawayprisoner

Uhh... proof? Again, the VMX units are for calculating floating point numbers only. Most games would depend on integer performance rather than floating point (unless it's physics calculations) so again, they are not going to be miraculous boosters. Even then, there are no data on VMX128 versus VMX performance so there's no telling if one core of Xenon is faster than the PPE or not.

They aren't use for just floating point numbers. The VMX unit is a combined SIMD and floating point unit. Even in the very link you posted it shows that.

You just said it was irrelevant to gaming.

I over exaggerated. But it is still true that general purpose computing has a very small impact to performance in games.

...If you say that to a developer, he'll smack you upside down the head. Really. General purpose codes take just as many cycles as physics and anything related to graphics to execute. Actually, sometimes even more cycles. Things like iterating through an array or looking up a specific symbol within an array are very cycle-intensive. It's mostly just sorting math when it comes to graphics related things (this is only to take off a bit of burden for the GPU).

Uhh... check your facts again, please. AI is not very easy to do on a weak processor. If you're talking about very simple AI (like in the asteroid types of games), then you're right. But if you're talking about advanced AI, then those are extremely CPU-heavy. And depending on the complexity of the game engine, the AI has to calculate a lot more. This shows the most in FPS games where the AI has to calculate the path that it takes as well as the trajectory of its shots and everything else related to that. About 10 of those things visible on screen and the burden increases ten-fold.

Oh Really?

Reviews & Previews | Gamekings - Elke dag nieuwe video's

O this video is very long. It's a 40 minute behind the scenes of how Killzone 2 was developed.

Here is some debug stats on how much of each part of the ps3 is being used. This was taken from the video.


CPU TIME
--------
Unknown .......... 1.24%
SPU Sync ......... 0.06%
AI Manager ....... 0.47%
Game Logic ....... 9.52%
Script ........... 0.80%
Physics .......... 1.57%
Representation ... 10.46%
Draw ............. 20.18%
HUD .............. 2.19%
Sound ............ 0.65%
Profile HUD ...... 25.17%
GPU Sync ......... 37.99%
----------
Total Time ....... 36.85%


SPU TIME
----------------------------
AI.Cover ................... ........ 0.00%
AI.LineOfFire .............. ........ 0.00%
Anim.EdgeAnim .............. 33 ..... 2.01%
Anim.Skinning .............. 152 .... 30.68%
Gfx.DecalUpdate ............ 9 ...... 0.78%
Gfx.LightProbes ............ 396 .... 9.00%
Gfx.PB.DeferredSchedule .... 1 ...... 0.60%
Gfx.PB.Forward ............. 2 ...... 1.69%
Gfx.PB.Geometry ............ 1 ...... 18.67%
Gfx.PB.Lights .............. 1 ...... 0.66%
Gfx.PB.ShadowMap ........... 1 ...... 4.20%
Gfx.Particles.ManagerJob ... 1 ...... 3.14%
Gfx.Particles.UpdateJob .... 130 .... 12.33%
Gfx.Particles.VertexJob .... 70 ..... 20.64%
Gfx.Post.BloomCapture ...... 12 ..... 2.80%
Gfx.Post.BloomIntegrate .... 8 ...... 1.52%
Gfx.Post.DepthOfField ...... 64 ..... 12.12%
Gfx.Post.DepthToFuzzy ...... 8 ...... 0.67%
Gfx.Post.Downsample ........ 29 ..... 0.61%
Gfx.Post.GrainWeight ....... 1 ...... 0.51%
Gfx.Post.HBlur ............. 45 ..... 3.02%
Gfx.Post.ILR ............... 1 ...... 0.63%
Gfx.Post.Modulate .......... 27 ..... 1.3?%
Gfx.Post.MotionBlur ........ 46 ..... 11.31%
Gfx.Post.Unlock? ........... 1 ...... 0.01%
Gfx.Post.Upsample .......... 108 .... 9.47%
Gfx.Post.VBlur ............. 46 ..... 3.73%
Gfx.Post.Vg??lle ........... 1 ...... 1.18%
Gfx.Post.Zero .............. 16 ..... 0.64%
Gfx.Scene.Portals .......... 3 ...... 30.72%
Mesh.Decompression ......... ........ 0.00%
Physics.Collide ............ 4 ...... 2.48%
Physics.Integrate .......... 4 ...... 2.11%
Physics.KdTree ............. 8 ...... 20.50%
Physics.Raycast ............ ........ 0.00%
Snd.MP3.Stereo ............. 2 ...... 2.60%
Snd.MP3.Surround ........... 2 ...... 7.51%
Snd.?Synth ................. 35 ..... 3.23%
Snd.Reverb ................. 14 ..... 4.02%
----------------------------
Total Time ................. 1232 ... 227.46%


GRAPHICS
--------
FPS ................. 30
GPU Stall by CPU .... 0.123 ?s
CPU stall by GPU .... 12.231 ?s


GPU TIME
--------------------------
Unknown ....... 0.2?/ ... 3.43%
Geometry ....... 1.8?/ ... 43.37%
Lighting ....... 1.7?/ ... 14.??%
Effects ........ 8.5?/ ... 8.4?%
Post process ............. 18.31%
--------------------------
Total Time ............... 81.??%
GPU Stall ................ 0.??%


PRIMS / TRI
-----------
Totals ..... 1431/ ... 344,634
Prime? ..... 0/ ...... 0
Geometry ... 619/ .... 161,231
Shadow ..... 683/ .... 170,???
Effects .... 121/ .... 14,3??


MEMORY STATS
------------------------
Pushbuffer ???? ........ 0.15 MB
Pushbuffer High ........ 0.15 MB
VRAM Free .............. 23.43 MB
Host Free .............. 80.?? MB
Heap Free .............. 134.?? MB
Render Mem ???? ........ 0
Render Mem Used ........ 12.00 MB
Render Mem Watermark ... 12.00 MB


MAIN RAM ....... 101.00 MB
----------------
Physics ........ 5.30 MB
Collision ...... 3.72 MB
Sound .......... 16.25 MB
Mesh ........... 21.20 MB
Graphics ....... 6.53 MB
Animation ...... 34.45 MB
Texture ........ 0.56 MB
Shader ......... 1.46 MB
AI Data ........ 2.75 MB
Various ........ 3.32 MB
Waste .......... 5.27 MB
----------------
Total .......... 97.17 MB
Main RAM ??? ... 97 / 101


VIDEO RAM .. 190.04 MB
------------
Mesh ....... 15.99 MB
Texture .... 156.87 MB
Waste ...... 1.20 MB
Total ...... 174.08 MB


Okay to explain something with the spe percentage and the ppe percentage. The total ppe percentage is 200%. 100% for each hardware thread. For the spes it is 700%. 100% for each spu. So the spu saying 227% means that they are using a little more than 2 entire spes.

Now Killzone 2 is a game known for it's complex AI. Look at how little cpu time AI takes up on the ps3. Also look at how the AI compares with the graphics and physics and animations.

If AI calculation was so easy, then Crysis and Far Cry 2 wouldn't get a boost from a faster CPU even though Physics were turned off.

Every game gets in performance when you turn the settings down to low settings and a low resolution. Not just Crysis and Far Cry 2.

That's great... but... the SPEs are still not full-fledged CPUs. They don't act like that. The most common mistake anyone can make is by thinking that they are CPUs. They act more like co-processors. If they were CPUs, then you could just bypass the PPE and get to the SPEs directly. Unless I'm missing something.

Okay. So they are not full fledged cpus but they are coprocessors. Happy now? This argument was all semantics anyway and it doesn't change my point.

Also the same Wikipedia said earlier that the Xenon had 2 vector units per core. I would take Wikipedia with a 50/50 mindset and check its sources before I make statements. Even the main Cell BE article on Wikipedia does not mention the SPEs as being vector processors.

It didn't say it had 2 vector units per core. I misread it. It has 2 sets of 128 registers per core. 1 set for each hardware thread.

Also, the article about the Cell broadband engine clearly says that the spes are SIMD units.

What will they make use of the VMX extensions for...? Even in games, they are not always necessary.

This is like saying that the SPEs aren't necessary and that game developers should not use them. You don't HAVE to use the VMX units, but if you want to make the most of your game you should use it.

The GFLOPS of the PS3 performing single-precision floating point calculations is up to 200. Yeah, that's great. But did you know that if you want to do precise physics calculations, you use double-precision floating point calculations? And when you do so, even the mighty Cell BE can only reach around 20GFLOPS or so. Here's a wikipedia quote since you like to check that place so much:

Cell (microprocessor) - Wikipedia, the free encyclopedia

And here's a quote, again, from Wikipedia:

FLOPS - Wikipedia, the free encyclopedia

Your fastest Core i7 can reach 70GFLOPS in double precision. It that regard, it's actually 3 times faster than Cell BE.

So? 3D gaming doesn't use or need double precision math. Single precision would be much better to use since 64bit(double precision) data takes up 2 times as much memory, cache etc. as 32bit(single precision) data. It's true that you can get more done at once with a 64bit word. But in the end most games simply don't use double precision floating points. Another downer is that todays processors process double precision math slower than single precision. Show me just one game that heavily uses double precision floating points. Now, if most games did heavily use double precision floating point numbers then yeah the i7 would easily trash the Cell processor in games. But they don't. It's not needed at this point in time.

The PPE in the Cell BE is rated for 6.4GFLOPS when it comes to double precision. So that's about 19.2GFLOPS with the Xenon. That is actually about as many GFLOPS as a 2.4GHz Core 2 Duo can dish out. Not mentioning a Core 2 Quad, of course. And not to mention the Core 2 Duo will smoke Xenon in general purpose tasks any day.

General purpose performance doesn't matter. Double precision doesn't matter at this point in time.

Actually, the ONE thing that drove Valve away from further releasing Source games on the PS3 was because the CPU could not keep up with the engine. Games like Left 4 Dead (up to 30 visible zombie AIs at the same time) would be almost impossible to do, and even on the 360, such games can only do 30fps at most.

Please check your facts.

Wait. Left 4 Dead is impossible to do on the PS3? WHAT! good joke you have right there. Valve doesn't make games on the ps3 because it is too hard for them. They aren't used to the ps3s architecture and they are more based on the PC. The Xbox 360's design is a lot more similar to our pcs than the ps3 and in general it is a lot easier.

http://loot-ninja.com/2009/06/09/val...mplicated-ps3/


The PC and the 360 are just more straightforward. We can focus on what we want to do, which is make game experiences, instead of sweating bullets over obscure architectural decisions they make with their platform. [...] I didn’t come into this business in the 90s because of some technical fetish. I came in because I wanted to give people experiences that made them have fun.

They specifically say that the ps3 is harder to deal with. There are plenty of games on the ps3 that beat left 4 dead 2.

I said that they don't do much. At least not to the point where you are trying to make them out to be. Physics calculations require very precise calculations. Most of the time with double precision float. That's why you don't see the likes of PhysX or Havoc on consoles as you do on PC. Xenon or Cell BE, whatever, are really really horrible at double-precision floating point operations. The situation is somewhat better with Cell BE and we can at least get performance close to PC, but it's almost a lost cause with 360.

Look at Crysis 2 on the 360. 'nuff said. And like I said before, double precision isn't used in games. Almost all games on the pc are 32bit. There are only a few games on the pc that are 64bit like Half Life 2 and Crysis. Crysis in 32bit mode has the same performance as 64bit, (actually slightly lower for 64bit according to some benchmarks), and 64bit Half Life 2 is not only slower but it is buggy. Either way we see that at this point in time, Double precision doesn't matter.

As for 3D graphics, it's only good if the CPU is what's rendering the scene. Unfortunately, the rendering is done on the GPU so the rest of the instruction extensions just seem really redundant.

That would be like saying that rendering on the spes are redundant.

Yeah, yeah... BUT, oh my God... what do you think VMX is used for? No, wait, actually, quote a game developer who says that the VMX instruction set in the Xenon works miracles.

Game Developers on the 360 in general don't talk much about the 360 development. They also don't seem to really push the hardware as much as ps3 developers. Most of the 360s big games this year were multi platform games. And most of the 360 exclusives are 3rd party games that would normally be multi platform. Sony has many more ambitous first party developers for the ps3 that make exclusives for the ps3 and actually try their hardest to get the most out of the system. Look at Heavy Rain and Killzone 2. The AI in those games are probably more complex than anything the 360 has, yet technically, the 360 has 3 times the general purpose processors and the 360 should be the system with more complex games in terms of AI. Especially since you say that AI has a big impact in performance.

No the Wikipedia page is right. I misunderstood. There aren't 2 vmx units per core. I thought that there were though. The diagram clearly shows 2*128*128bit registers in the VRF(the register files) part of the VMX 128 unit. So basically 2 sets of 128 128bit registers. I assumed it meant VMX units. I assumed it meant 2 vmx units when wikipedia showed 2*128*128bit. My bad. My point still stands though. A Single vmx unit easily out performs the SSE units in a Core 2 duo @ 2.0GHz.



The VMX unit is a combined SIMD(vector) and FPU unit. You can't just simplfy it to just an FPU. That would be like calling the Cell processor a general purpose processor.

I posted that and you didn't respond to it. You just reposted it without quotes.

...Uhmmmm... wow. Do you know that if you use the CPU to render 3D graphics then you have to use the CPU to complete at least that rendering task before sending it to the GPU? That's dead slow. No, wait, that's not just dead-slow, that's crawling. You need to link up 2 or more PS3s to be able to get around 30fps rendering complete scenes all by themselves at game resolutions. How "fast" do you think the Xenon CPU is to be able to render 3D graphics along with the GPU?

Yes that is if you use the ps3 cpu by itself. I specifically said you need to use the cpu and the gpu together. In Killzone 2, they do the entire task of Deferred rendering on the spus. Uncharted 2 uses even more of the spes. The Crytek says that they are using the spus the most out of any developer out now and it shows since Cryengine 3 on the ps3 looks way better than what a 7800 series gpu would be able to do(The gpu in the ps3 is a modified g70)

Read above, please. Single-precision math, it may be faster. But double-precision math... erg.

double precision math doesn't matter and at this point in time... erg

Again, these things are not miracle workers. Granted, Core 2 is not that much faster if they have to do single-precision math

The Core 2 is not faster at all with single precision. Actually, the core 2 is many times slower with single precision.

but double-precision math is more used in physics simulations because it's more accurate that way

Double precision is more accurate, but it is not needed, and not used at this point in time. Single precision is accurate enough.

and both Cell BE and Xenon take humongous (up to 90% performance hit) when calculating double-precision. That's why developers keep saying PS3 and 360 perform similarly and 360 has a better GPU (Cell BE and Xenon both have around 20GFLOPS of performance when it comes to double-precision) even though we keep seeing Cell BE being 3 times faster than Xenon.

You know what else takes a huge 90% hit in double precision performance? Nvidia gpus. The 8800 and 9800 series don't even SUPPORT double precision floating points. You need at least a GTX 260 to even get support for double precision, and even then the performance is extremely low. Yet we see them being used with PhysX. Can you guess why? Because right now for gaming double precision doesn't matter, it does matter on for hpc markets and supercomputing depending on what the user needs, but for gaming it is not needed right now.

Eh... you forgot the fact that Xenon is based on in-order processing. The Dashboard might be light, but it's not so light that the processor doesn't take some hits from it.

It's light enough to say that it doesn't matter in this specific discussion. The Intel Atom can handle running Windows95 easily. The atom is many times weaker than the Xenon, and Windows 95 is heavier than the Dashboard.

Turn even HyperThreading off then.

Without hyperthreading of course gta 4 will run worse. It is still playable. But it is below the ps3 and xbox 360 performance.

No, a Core 2 Duo @ 2GHz is equal to a Core i7 without HyperThreading running on 2 cores at 1.93GHz.

Nope. Clock for clock the i7 is much faster than that.

AnandTech Bench (beta): Intel Core i5 750 vs Intel Core 2 Quad Q9400

You can even look at some of the cpu benchmarks I posted before.

So you do 1080p encoding while you play a game?

You said that vector processors are maybe better at some forms of multimedia. I was just posting as many examples to show how you were trying to downplay the advantage a vector processor has over a Superscalar processor. .

That's point one. On PC, you can just throw everything onto the GPU and be done with it (what CryEngine 2 does). That's the mindset.

Isn't that what I've been saying.

Uh... they don't just have AI problems with the Cell BE. They have ALL sorts of problems with the Cell BE. Technically, they have problems processing everything, not just AI. AI plays a part, though.

Nope.

If not then Valve wouldn't complain and quit the PS3. The design is not more complex than a Core 2 Quad or anything like that.

They quit the ps3 because it is too hard and complicated form them. They aren't good enough programmers to make use of the ps3. They said THEMSELVES that the pc was a lot more straightfoward and that the ps3 is more complicated and harder to deal with.

And it is an absolute bold faced lie to say that the Core 2 quad is as complex as the Cell. The Core 2 quad features just 2 Core 2 Duos strapped together and it is just a supercalar processor.

The Cell processor is a super scalar processor and a vector processor at the same time. That alone proves my point. Hell I'm shouldn't even explain this since Valve themselves said it was the reason they are not making games for the ps3 right now.

The problem is that the CPU is too slow at some tasks that would otherwise run very fast on some other systems. But they can't just say that the Cell BE because then they'd get laughed at, so they just chalk it off as complex design.

They went and said that they aren't good enough programmers for the ps3 and that they are too lazy to program for the ps3. That would be a lot worse for them than saying the ps3 is slow.

Uhh... there's no code "optimized" for out-of-order or in-order execution. They are just different execution style. Obviously... very compact codes will run fast in both, while very heavy codes will run slower on in-order. It's still up to developers. In some cases, you just can't avoid heavy codes.

If you improve the parallelism of code on an instruction level, then the performance impact of in order processors will be reduced. If a compiler is coded can do this well then the performance impact wouldn't be as large.

It's not what I'm doing. It's what developers do. And if they've done as you've said that they should do, then you'd see consoles outperforming PCs by miles when it comes to physics. They don't. And I put the reason above.

And we should be seeing a dual QX9775 skulltrail system easily beating a Core 2 Quad in todays games. But it doesn't. It is barely stronger than a Core 2 Duo e8600 in games like Crysis. I guess 2 QX9775 cpus aren't much stronger than a single core 2 quad or a core 2 duo. That has to be it, it can't possibly be that developers aren't making use of the extra power.

There is only one set. Because VMX 128 can only work with 128 registers at the same time. What the diagram meant to say was 2 threads, both able to access the common 128 registers at 128-bit length for each register.

There are 2 threads, but there are 128 registers for each hardware thread look it up.

- The CPUs are horrible at double-precision maths, which are most often used for physics simulations.

lol often used. More like barely used.

They're not that horrible at single-precision, but theoretical performance =/= practical performance.

So what, if a cpu has over 100GFlops of theoretical performance, it will most likely easily beat a cpu that has 16GFlops of theoretical performance in real world tests. Plus, like I said before apple was able to get nearly 100% of the theoretical performance out of the VMX units.

- VMX instructions are cool and all, but if multicore support from developer sucks then performance also suffers. The past years have shown that for us. Even with graphics chips with thousands of shader processors, efficiency is not at 100%. If they were then HD4870 would smoke GTX 285 by miles

.

AMD and Nvidia count their shaders differently/ If AMD counted their shader processors the same way Nvidia counts theirs then AMD would be saying that they have way way less shaders than what they say now. PLUS their cards use different architectures so the comparison is a bad one.

The real world performance difference between an 800 shader core and a 400 shader core from AMD using the same architecture is nearly linear.

- 3D rendering is not done on the CPU... at least not on Xenon because the CPU is not fast enough for that.On the PS3, the SPEs only do the preparing work. The rendering still has to be done on the GPU. That's how it's done on PC, too, though.

No. On the ps3, in Uncharted 2, all of the post processing effects are done with the spes. Same for Killzone 2. With Killzone 2, they do deferred rendering with the spes and not the gpu. On the ps3, the makers of the sabetour used the spes to do 16x Anti aliasing. That is not the way developers do things on the pc.

- A Core i7 will outperform Cell BE or Xenon at double-precision by many folds. That's because double-precision counts more than single-precision.

No, double precision doesn't matter. I'm surprised that you would even bring up double precision performance in the first place. It is totally irrelevant.

- Just a fun note: Flash also runs horrible on the PS3 even though it's touted as this very fast CPU. Simply because all Flash maths involve double-precision.

What does Flash have to do with gaming?

[makotech222]

lol left 4 dead on the xbox360 is HORRIBLE. It runs at like 20 fps all the time. A world of difference from my PC which plays it easy at 120 fps, with better graphics, and # of zombies on screen.

__________________

Desktop: [Windows 7 x64][Intel E8400@3.8ghz] [DDR2 4gb] [Nvidia EVGA Gtx 260] [Creative X-Fi Soundblaster] [Corsair 550w][Viewsonic 23inch 120hz Display]
Lenovo X200Tablet: [Windows 7 build 7600 x86][L9400 @1.86ghz][Intel GMA 4500MHD][2gb Ram][160GB 7200rpm][LED-lit Touchscreen/pen-abled]

[runawayprisoner]

Originally Posted by bkwegoharder

They aren't use for just floating point numbers. The VMX unit is a combined SIMD and floating point unit. Even in the very link you posted it shows that.

...a floating point unit that's not just used for calculating floating point numbers. What are you trying to say...?

SIMD is a technique... or more like instruction set or more like a design of the instruction set rather than a processor type. Urg... even Wikipedia said that.

I over exaggerated. But it is still true that general purpose computing has a very small impact to performance in games.

How is it true... please do tell. I can list over thousands of ways that games are more general-purpose than your general-purpose tasks.

Oh Really?

Really.

Okay to explain something with the spe percentage and the ppe percentage. The total ppe percentage is 200%. 100% for each hardware thread. For the spes it is 700%. 100% for each spu. So the spu saying 227% means that they are using a little more than 2 entire spes.

Uhmm... 227% means total SPU usage. Which could be all 7 SPUs each being used up to 30%. Problem here is that you can't push data and schedule them so perfectly that 100% of a SPU can be used. You can't tell how much time it would take for one SPU to calculate a certain type of data, and all you can do is pretty much... see that if one SPU is busy calculating one task then you get to the next SPU and throw the task at it if it's free, then cycle around after making sure that data are being calculated in the correct order or they go where they should be. Consult PCSX2 developers on why it has to be that way.

Please... study more programming before you make statements like this. I don't want to sound like an *** but really, if you say this to someone who really knows his stuffs, the lightest you'll get out of it is a laugh from the guy. I presented this data the same way to a certain developer and got laughed at... bad.

Now Killzone 2 is a game known for it's complex AI. Look at how little cpu time AI takes up on the ps3. Also look at how the AI compares with the graphics and physics and animations.

Killzone 2 AI is not complex at all. All it does is hide behind covers, shoot blindly. If you get close, it charges at you, if you shoot at it for too long, it runs to another cover.

And there's hardly that many AI always visible on the screen. Try Left 4 Dead. That game has much more complex AI. And Valve pretty much chalked it off as impossible on the PS3.

Every game gets in performance when you turn the settings down to low settings and a low resolution. Not just Crysis and Far Cry 2.

Uhh... but they get higher performance when you use a faster CPU. No? Lowest settings means they can only be CPU-bound. And fact is... they are.

Okay. So they are not full fledged cpus but they are coprocessors. Happy now? This argument was all semantics anyway and it doesn't change my point.

It changes everything actually.

It didn't say it had 2 vector units per core. I misread it. It has 2 sets of 128 registers per core. 1 set for each hardware thread.

It's a common set. You have two threads but they both access the same register set. The reason is because it's not two FPUs executing 2 threads. It's only 1 FPU. With 2 threads.

Also, the article about the Cell broadband engine clearly says that the spes are SIMD units.

...Please read again the definition of SIMD.

This is like saying that the SPEs aren't necessary and that game developers should not use them. You don't HAVE to use the VMX units, but if you want to make the most of your game you should use it.

No, I'm not saying that. I'm saying that the VMX units either on the PPE or on each core of the Xenon, no matter how improvised, don't provide that much of a boost as you are making them out to be. And the SPEs are not just for floating point calculations, but they are not full-fledged CPU or vector processors either. They are coprocessors, like you said, but that means their operations mostly depend on how fast the PPE can react, and that's why some games on the PS3 run slower than on the 360. That's just a fact. The PPE is THAT slow and there's no magical cure that either Sony or Microsoft put it to remedy the situation.

So? 3D gaming doesn't use or need double precision math. Single precision would be much better to use since 64bit(double precision) data takes up 2 times as much memory, cache etc. as 32bit(single precision) data. It's true that you can get more done at once with a 64bit word. But in the end most games simply don't use double precision floating points. Another downer is that todays processors process double precision math slower than single precision. Show me just one game that heavily uses double precision floating points. Now, if most games did heavily use double precision floating point numbers then yeah the i7 would easily trash the Cell processor in games. But they don't. It's not needed at this point in time.

Uhh... actually most games that do physics will use double precision floating point because they are more precise and you can avoid lots of hit test errors that way. If you use single precision then what will happen is that sometimes your character "sinks" into the ground or the likes. And single-precision math can be substituted with very large integers anyway. And integer operations are much much much faster than floating point operations.

Does the i7 trash the Cell processor in games...? Uhh... actually, it does. You're basically saying that PC games are inferior to console games.

Well... seriously, I'll just take for example... Street Fighter IV. Slap a GeForce 7600GT onto a Core i7 platform, and then run the same game. I'll bet anything that you'll get significantly better performance compared to the same game running on the PS3. Try the same thing for Far Cry 2, too if you have the chance.

General purpose performance doesn't matter. Double precision doesn't matter at this point in time.

...you're really going into denial mode, aren't you? Try asking around why double-precision is so important. I really can't be arsed to make this text any longer just to show you why.

Wait. Left 4 Dead is impossible to do on the PS3? WHAT! good joke you have right there. Valve doesn't make games on the ps3 because it is too hard for them. They aren't used to the ps3s architecture and they are more based on the PC. The Xbox 360's design is a lot more similar to our pcs than the ps3 and in general it is a lot easier.

Valve said it was impossible. Go by their words. Did you code for the PS3? I didn't think so.

They specifically say that the ps3 is harder to deal with. There are plenty of games on the ps3 that beat left 4 dead 2.

Find one game on the PS3 that has 30 enemies on the screen at once, each reacting to your every movement, even the sounds of your footsteps.

I can't find one... honestly. Not even Resistance or Killzone 2 have that kind of AI or that many enemies on the screen at once. You'd think Left 4 Dead AI is simple, but if you play the game on PC, you'll see. It's not just simply zombies charging at you.

Look at Crysis 2 on the 360. 'nuff said.

Yeah... 30fps in the best case scenario, 18-20fps when there are lots of **** on the screen at once and not even a single AI in the scene just yet. Even worse on the PS3. Significantly reduced draw distance, too. A barrel only materializes when the player gets to about 10 feet radius from the base.

And like I said before, double precision isn't used in games. Almost all games on the pc are 32bit. There are only a few games on the pc that are 64bit like Half Life 2 and Crysis. Crysis in 32bit mode has the same performance as 64bit, (actually slightly lower for 64bit according to some benchmarks), and 64bit Half Life 2 is not only slower but it is buggy. Either way we see that at this point in time, Double precision doesn't matter.

Oh my God...

Double-precision means the data is 64-bit long. It doesn't mean it's only possible when the processor is 64-bit capable.

Variables. Data Types.

There's even a long double primitive type in C++ that's 128 bits long. That doesn't mean a 32-bit processor can't process it. Heck, sometimes the data type you want to process can be even longer. 256 bits or even 512 bits. A 32-bit processor will be able to calculate that just fine. Will a 64-bit processor be able to calculate 64-bit data types faster? Likely, but not always.

Game Developers on the 360 in general don't talk much about the 360 development. They also don't seem to really push the hardware as much as ps3 developers. Most of the 360s big games this year were multi platform games. And most of the 360 exclusives are 3rd party games that would normally be multi platform. Sony has many more ambitous first party developers for the ps3 that make exclusives for the ps3 and actually try their hardest to get the most out of the system. Look at Heavy Rain and Killzone 2. The AI in those games are probably more complex than anything the 360 has, yet technically, the 360 has 3 times the general purpose processors and the 360 should be the system with more complex games in terms of AI. Especially since you say that AI has a big impact in performance.

And... PS3 doesn't have Left 4 Dead. Enough said.

Some games on the PS3 also lack in visuals and resolutions. Typically down to shadows and such. Those things need to be calculated by the CPU, too.

Yes that is if you use the ps3 cpu by itself. I specifically said you need to use the cpu and the gpu together. In Killzone 2, they do the entire task of Deferred rendering on the spus. Uncharted 2 uses even more of the spes. The Crytek says that they are using the spus the most out of any developer out now and it shows since Cryengine 3 on the ps3 looks way better than what a 7800 series gpu would be able to do(The gpu in the ps3 is a modified g70)

Uhh... set object quality to low in Crysis then bump everything else to High or Medium. I'm sure a 7800GTX won't have any problem dealing with that at over 30fps. Or in fact, since I have a 7800GTX here, I'll try it myself during the weekend.

double precision math doesn't matter and at this point in time... erg

Um... I can't believe I'm saying this... but please go learn programming.

The Core 2 is not faster at all with single precision. Actually, the core 2 is many times slower with single precision.

There's no denying that, but games would use double-precision if it's physics or hit tests.

Double precision is more accurate, but it is not needed, and not used at this point in time. Single precision is accurate enough.

Single precision is not accurate enough. It's only as accurate as you using very large integers because the limit is still 32-bit. You can't store anything longer than that in there.

That's why double-precision exists... and even long double exists. They need things to be more accurate.

You know what else takes a huge 90% hit in double precision performance? Nvidia gpus. The 8800 and 9800 series don't even SUPPORT double precision floating points. You need at least a GTX 260 to even get support for double precision, and even then the performance is extremely low. Yet we see them being used with PhysX. Can you guess why? Because right now for gaming double precision doesn't matter, it does matter on for hpc markets and supercomputing depending on what the user needs, but for gaming it is not needed right now.

You can emulate double-precision math. They just don't have hardware support for double-precision math. It doesn't mean you can't do it in software.

PhysX is indeed not in double-precision, though, and that's why it remains as something that only calculates particle effects or physics effects that don't actually require a lot of precision. And people keep wondering why they don't see buildings crumbling and destructible walls and the likes. It's just not precise enough for that task.

And that's also why Little Big Planet on the PS3 doesn't have fluid clothes simulation like in the demo... and a few other things were also taken out. It would have glitched the game to death if they used single precision all the way. Granted... single-precision DOES work in some cases, but it'll cause glitches.

It's light enough to say that it doesn't matter in this specific discussion. The Intel Atom can handle running Windows95 easily. The atom is many times weaker than the Xenon, and Windows 95 is heavier than the Dashboard.

Try stripping Windows 7 down with vLite and use Intel Atom with it. Then try stripping Windows XP down with nLite and use Atom with it. Huge difference in performance.

And I'm not sure if Atom is weaker than Xenon. It's weak, but not by many times. At most half that the performance of Xenon.

Without hyperthreading of course gta 4 will run worse. It is still playable. But it is below the ps3 and xbox 360 performance.

Even with 2 cores enabled? Huh...

Well, I won't go quoting more from this point on. Please... again, try to learn programming. Or at least try to look more into programming. You're discussing something that involves programming without any concrete backup at all.

[bkwegoharder]

Originally Posted by runawayprisoner

...a floating point unit that's not just used for calculating floating point numbers. What are you trying to say...?

SIMD is a technique... or more like instruction set or more like a design of the instruction set rather than a processor type. Urg... even Wikipedia said that.

What are you talking about? Even ATI calls each unit of their gpu a SIMD unit. What do you think a SIMD unit processes? SIMD instructions of course. What you are saying is like saying an x86 is just an instruction set and there is no such thing as an x86 processor.

How is it true... please do tell. I can list over thousands of ways that games are more general-purpose than your general-purpose tasks.

Name them. Then explain how they have a bigger impact to performance than the 3D graphics, Physics, Particle effects, animations, audio and all of the other things games do that don't have to do with General purpose processing. Then explain how they are more general purpose than running an OS with a lot of different random applications running.

Uhmm... 227% means total SPU usage. Which could be all 7 SPUs each being used up to 30%.

The developers themselves said specifically at 18:46 in the 40 minute video that I posted that normally they use 2 of the six spus. I simply mirrored what they said. I guess they are wrong though since you know more than actual PS3 developers.

Also, as a side note they were able to do things 20%-40% faster using instead of using the graphics card. An i7 will never be able to do graphics 20%-40% faster than a 7800GTX. Even a very weak gpu like the 8600GT will easily mop the floor with the i7 when it comes to 3D graphics simply because the i7 is not a processor made to do 3D graphics. The Cell processor wipes the floor with it easily, in 3D graphics and the other things I said. They simply don't compare and trying to argue otherwise is foolish. Likewise when it comes to general purpose computing and multitasking the i7 wins easily like I said before.

Consult PCSX2 developers on why it has to be that way.

Oh you want to play that game? Okay Consult a pcsx2 developer about how the i7 pales in comparison to the Cell processor for gaming. wait, no I did it for you already.

Sony's plans for PS2 and Dreamcast emulation on PS3

If you want to talk "gaming performance" then none of my above comparisons are relevant. When playing games natively (not emulated), 75-90% of the work is done by the Cell's 7 SPEs, depending on game. When emulating a PS2, those will be largely unutilized.

The SPE design is a very good one for gamedev in general. It'll murder the i7+GX290 without breaking a sweat, as far as gaming is concerned, assuming the programmers have a strong grasp of parallel programming concepts (which Sony coders certainly do). There are some other "flatter" multimedia tasks where the GX290 can hold its own against the Cell, but gamedev usually benefits nicely from the fact that SPEs are complete co-processors, and not just souped-up pixel shaders. (which is why the future generations of GPUs will most likely start moving in that direction)

This whole argument I could have just posted that, but I decided not to because I wanted to support my argument with my own words without copy & pasting what someone else said. But I guess this lays it all to rest. The Cell murders the i7 in gaming. Point blank period. End of discussion.


EDIT: The PCSX2 is down right now so here is a cached snapshot from google of what I was talking about

http://74.125.93.132/search?q=cache:...ient=firefox-a

Killzone 2 AI is not complex at all. All it does is hide behind covers, shoot blindly. If you get close, it charges at you, if you shoot at it for too long, it runs to another cover.

And there's hardly that many AI always visible on the screen. Try Left 4 Dead. That game has much more complex AI. And Valve pretty much chalked it off as impossible on the PS3.

The AI is more complex than Left 4 dead believe that. It is not only about how the AI combats you but it is about how they react to you. If you shoot a guy in the arm, he'll actually react like he got hit in the arm, unlike in left 4 dead. In Left 4 dead, the zombies rush at you, you shoot them, they die and drop dead to the floor. GTA 4 not only has the elements of Killzone 2 where the people in it react realistically, they also have what Left 4 Dead has where there are lots and lots of people on screen.

Steal a bus, and park them in Time square across the intersections to cause a traffic jam. Take out your AK 47, and then start shooting at the sidewalk full of people, they will all start running away from you, all of the cars will start to go out of control scrambling to get away from the carnage while you are shooting the sidewalk full of pedestrians. One of the pedestrians you hit got hit in the arm. He starts holding his arm in pain. Then he looks at where you shot him, then he looks straight at you in utter fear with a terrified look on his face. Another guy gets hit from the bullets while he was running, he gets hit in the leg. He falls over and hits the ground realistically. then he struggles to get up and he starts limping away in fear. The police come and they all are trying their hardest to shoot you down to save the people who haven't been killed yet. All of this happening all at once. Lot's of people on screen, lots of things happening at once. And in general it is just a very very horrific and sad columbine style massacre.

That is way more than Left 4 Dead is, it is more than it ever was and it is more than it probably ever will be. There is no excuse for why Left 4 Dead isn't on the ps3 other than Valve simply not knowing how to program correctly for the ps3 and them just not wanting to. There is nothing wrong with that though. If they can't do it then it is there perogative(I know I spelled that wrong). They said specifically that they wanted to simply make fun games without going through hoops of fire to do it, and that is all good.

They specifically said that it was too complicated and that they didn't want to put in the effort. I don't think that they said that Left 4 dead can't be done on the ps3. Can you post a link to them saying that it can't be done?

Even if they said that it can't be done, I don't think that they would be an authority on what the ps3 can and can't do since they actually never made a game for the ps3. The Orange Box on the ps3 was was made by a different company according to what they said.

Uhh... but they get higher performance when you use a faster CPU. No? Lowest settings means they can only be CPU-bound. And fact is... they are.

Nice cop out.

By that logic GTA 4 is gpu limited because I run it at 2560 x 1600 and moving from a 9600GT to a GTX 280 greatly improves my performance.

I was clearly talking about typical settings on a gaming pc today. Of course a game is going to be cpu limited when running a game at 800 x 600 at low settings. All games are.

It's a common set. You have two threads but they both access the same register set. The reason is because it's not two FPUs executing 2 threads. It's only 1 FPU. With 2 threads.

It's one SIMD unit. With 128 registers per hardware thread. The 360 has 6 hardware threads so in all there are 768 registers. Search google. Matter a fact here.

Xbox 360 128 registers per hardware thread - Google Search

I did it for you. You don't get just one website but many websites.

...Please read again the definition of SIMD.

Why don't you?

No, I'm not saying that. I'm saying that the VMX units either on the PPE or on each core of the Xenon, no matter how improvised, don't provide that much of a boost as you are making them out to be.

You aren't proving it. You are just saying that it can't happen. you haven't showed proof yet.

And the SPEs are not just for floating point calculations, but they are not full-fledged CPU or vector processors either. They are coprocessors, like you said, but that means their operations mostly depend on how fast the PPE can react

If you look at the Killzone 2 debug info, it shows you that the cpu is barely stalled by the spus. The PPE is good enough to take on the task.

and that's why some games on the PS3 run slower than on the 360. That's just a fact.

No, they are typically slower because the 360 is easier to program for and it is easier to get acceptable performance from it.

Uhh... actually most games that do physics will use double precision floating point because they are more precise and you can avoid lots of hit test errors that way. If you use single precision then what will happen is that sometimes your character "sinks" into the ground or the likes. And single-precision math can be substituted with very large integers anyway. And integer operations are much much much faster than floating point operations.

Crysis 2 runs just fine on the 360 and ps3 without having objects sink into the floor.

Does the i7 trash the Cell processor in games...? Uhh... actually, it does. You're basically saying that PC games are inferior to console games.

Nope. It doesn't. Show me an example of a pc gaming running faster on the pc with an i7 and a 7800GTX 256MB.

Well... seriously, I'll just take for example... Street Fighter IV. Slap a GeForce 7600GT onto a Core i7 platform, and then run the same game. I'll bet anything that you'll get significantly better performance compared to the same game running on the PS3. Try the same thing for Far Cry 2, too if you have the chance.

Why don't you just prove your point. i'm not going to get 2 versions of Street fighter 4 and a 7600GT to prove the point you made.

...you're really going into denial mode, aren't you? Try asking around why double-precision is so important. I really can't be arsed to make this text any longer just to show you why.

That is not how you make a statement and provide proof for your argument. You are supposed to present your argument, then PROVIDE FACTS and EXAMPLES proving your argument. You don't just make an argument then try and call the other person ignorant if they ask you for proof to back up your argument. but the fact is that you can't provide proof so you are trying to cop out by saying that you can't be arsed to make this text any longer. You did the same with the AI argument. You still haven't shown proof, you just said that a game developer would slap me upside my head if I said that ai doesn't have a big impact in performance. Okay, if that is true then tell me what game developer. I've shown you tons of info backing up my argument. You haven't shown proof yet though. What you like to do is make a statement, then act as if everyone knows the statement is true, then try to ridicule the person asking questions by saying things like "dude you need to look this up" or " do some research" etc. That is not how you win a debate.

Why don't you ask around on why AI does not take up many cpu cycles? I shouldn't have to type my text explaining all of this.

Here is what another person invovled with the development says about the ai thing.

Will GPU ever be more important than CPU?

The detail the media doesn't talk about here is that tasks like the AI are just not so demanding as to need a whole cpu core

Okay now this isn't as blatant as the first thing I posted from pcsx2.net. But it does support my argument.

EDIT: PCSX2 is down so here is a cached google snapshot

http://74.125.93.132/search?q=cache:...ient=firefox-a

Now you have what I said about the whole AI issue. What the pcsx2 developer said about how it is not so demanding. You have that other link I posted that said PS3 hardware explained with the guy saying ai and general purpose computing in general doesn't have a big effect on performance. You have the debug stats from Killzone 2 that showed a very low amount of cpu time being taken up by AI. You have what Air from pcsx2 said about the Cell murdering the i7 in gaming(If that is so then obviously general purpose computing doesn't matter a lot in gaming since the Cell is very very weak in terms of general purpose computing and the i7 is very very very strong in that regard).

And with all of that you still haven't shown one instance or game or an example of anyone saying that ai does matter a huge lot in gaming. Or of anyone saying that general purpose computing in gaming matters a lot. Instead of you constantly telling me to ask around and do research. Why don't you do some? Now I'm not trying to seem mean, but the things you say like "do your research" "Look this up" "ask around" etc. is getting on my nerve. You act as if I just pop up in this topic spouting a bunch of BS that I made up on the spot.

Valve said it was impossible. Go by their words. Did you code for the PS3? I didn't think so.

Valve haven't developed a game for the ps3 yet. their words mean nothing. Plus I don't even see where they said Left 4 Dead can't be done on the ps3. If they did say that though, it still doesn't matter because they have yet to develop a game for the ps3.

Find one game on the PS3 that has 30 enemies on the screen at once, each reacting to your every movement, even the sounds of your footsteps.

GTA 4 nuff said. Granted they don't react to the sounds you make, but all in all the characters in gta 4 are much more complex and there is a lot of things and a lot of people doing things at once.

I can't find one... honestly. Not even Resistance or Killzone 2 have that kind of AI or that many enemies on the screen at once. You'd think Left 4 Dead AI is simple, but if you play the game on PC, you'll see. It's not just simply zombies charging at you.

It's not on the level of GTA 4 or killzone 2. And even then, AI doesn't take away many cpu cycles.

Yeah... 30fps in the best case scenario, 18-20fps when there are lots of **** on the screen at once and not even a single AI in the scene just yet. Even worse on the PS3. Significantly reduced draw distance, too. A barrel only materializes when the player gets to about 10 feet radius from the base.

And the 7800GTX still won't manage to do all of what the Crysis 2 is on the consoles.

Oh my God...

Double-precision means the data is 64-bit long. It doesn't mean it's only possible when the processor is 64-bit capable.

Variables. Data Types.

There's even a long double primitive type in C++ that's 128 bits long. That doesn't mean a 32-bit processor can't process it. Heck, sometimes the data type you want to process can be even longer. 256 bits or even 512 bits. A 32-bit processor will be able to calculate that just fine. Will a 64-bit processor be able to calculate 64-bit data types faster?

I wasn't talking about a 32bit processor. I'm talking about the executable. Tell me how are you going to use a 64bit word in a 32bit executable?


Also, show me proof of games actually using double precision floating point numbers.

Likely, but not always.

Okay elaborate on this.

And... PS3 doesn't have Left 4 Dead. Enough said.

And the pc doesn't have Uncharted 2 so we all know that it obviously can't be done. M I RIGHT?

Uhh... set object quality to low in Crysis then bump everything else to High or Medium.


I'm sure a 7800GTX won't have any problem dealing with that at over 30fps. Or in fact, since I have a 7800GTX here, I'll try it myself during the weekend

Well we know that Object Quality will be set to low, shaders will be set to high because on medium it is obvious that the consoles are superior. Post Processing will have to be at very high for the sun rays, or if there is some way to get the sun rays in DX9 mode with one of those mods I remember seeing. Physics will have to be at atleast High.(Which is another thing, actually look at the physics effects on the consoles. And you say great physics can't be done on the consoles?)

There's no denying that, but games would use double-precision if it's physics or hit tests.

Single precision is not accurate enough. It's only as accurate as you using very large integers because the limit is still 32-bit. You can't store anything longer than that in there.

That's why double-precision exists... and even long double exists. They need things to be more accurate.

Okay show me a real world example of games using double precision floating points.

You can emulate double-precision math. They just don't have hardware support for double-precision math. It doesn't mean you can't do it in software.

Okay, so wouldn't Crysis get a performance boost instead of performance loss from using a 64bit executable by that logic?

PhysX is indeed not in double-precision, though, and that's why it remains as something that only calculates particle effects or physics effects that don't actually require a lot of precision. And people keep wondering why they don't see buildings crumbling and destructible walls and the likes. It's just not precise enough for that task.

And that's also why Little Big Planet on the PS3 doesn't have fluid clothes simulation like in the demo... and a few other things were also taken out. It would have glitched the game to death if they used single precision all the way. Granted... single-precision DOES work in some cases, but it'll cause glitches.

Okay, show me a link of proof or something about what you said about Little Big Planet. Also, explain to me why there a fluid cloths in Mirrors Edge using PhysX if it is not possible without glitches.

Try stripping Windows 7 down with vLite and use Intel Atom with it. Then try stripping Windows XP down with nLite and use Atom with it. Huge difference in performance.

There is a big difference between Windows 95 and Windows 7. And there is a big difference between the 360 dashboard and Windows95.

And I'm not sure if Atom is weaker than Xenon. It's weak, but not by many times. At most half that the performance of Xenon.

Nope. The Intel Atom is like half the strength of even a Pentium M @ 1.73GHz.

Intel Atom - Wikipedia, the free encyclopedia

Unless you are going to try and say that the Xenon is the same strength of a Pentium M now.

Even with 2 cores enabled? Huh...

No with 1 core enabled. I thought that I made it clear that I was running with a single core with hyperthreading. Disabling hyperthreading gives me just one core, and yes the game is playable if you lower the settings enough.

Well, I won't go quoting more from this point on. Please... again, try to learn programming. Or at least try to look more into programming. You're discussing something that involves programming without any concrete backup at all.

You are simply deciding to stop here because you can't admit you are wrong at some things.

Like saying the way games are rendered on the pc is done the same way as on the ps3. Gurellia Games used the CPU mainly to do everything our gpus would normally do, then they went to the gpu second. That is not how our PC games do things. I don't see any pc games doing anti aliasing with the cpu.






Also, more on the double precision thing. I almost forgot, but double precision support is actually something speculated to be cut out of Fermi for the gaming pc markets and average consumers. Why? Because double precision is for the most part not needed and not used and it would up the cost and heat, etc. of Fermi and no one would benefit from it.It is good for higher end markets, but it isn't really needed in gaming.

http://www.xbitlabs.com/news/video/d...ics_Cards.html

---

Last edited by bkwegoharder; December 25th, 2009 at 21:05..

[Krayzie B.o.n.e]

The Cell would indeed kill a core i7. Most games today off load all the work to the GPU. Your Cpu needs to only hold the environment together and A.I. the GPU does everything else. Do you actually think your Cpu can do Anti Ailising, run your OS, do water tessellation, monitor your firewall, and do explosions all at once? No your CPU would have a heart attack. High end graphics cards have GPU's that are more powerful than Cpu's.I t's just right now GPu's only do one thing and that's accelerate graphics.

Killzone 2 and Uncharted 2 have so far used the Cell the most. Using the PPE and two or three SPE's. Look how great those games are. Wait until someone uses the PPE and four or five SPE's properly, the game would be awesome. But A high end PC will still mash a PS3 because the video card in the PS3 is weak compared to a high end PC.

---

Last edited by Krayzie B.o.n.e; December 25th, 2009 at 06:43.. Reason: Automerged Doublepost

[runawayprisoner]

Originally Posted by bkwegoharder

Name them. Then explain how they have a bigger impact to performance than the 3D graphics, Physics, Particle effects, animations, audio and all of the other things games do that don't have to do with General purpose processing. Then explain how they are more general purpose than running an OS with a lot of different random applications running.

Physics are a bunch of "ifs" and "elses" and in most cases "switch...case..." conditions for hit tests. Then data is processed one way or another.

Z-order or occlusion culling is a sorting algorithm in most cases. You do the same thing with Excel cells here.

All game tasks are general-purpose one way or another. But it's not apparent to your general users because they don't know what is going on underneath. Try programming a game, then try programming an application, and you'll see that there's not much difference between the two. If anything, you calculate more, and use the GPU more. But that's all.

The developers themselves said specifically at 18:46 in the 40 minute video that I posted that normally they use 2 of the six spus. I simply mirrored what they said. I guess they are wrong though since you know more than actual PS3 developers.

Normally, doesn't mean they are only using 2 ALL THE TIME. You're just bending their words.

Also, as a side note they were able to do things 20%-40% faster using instead of using the graphics card. An i7 will never be able to do graphics 20%-40% faster than a 7800GTX.

20% - 40% faster offloading work onto the Cell BE and leaving the GPU open for other things. You're bending their words again.

Even a very weak gpu like the 8600GT will easily mop the floor with the i7 when it comes to 3D graphics simply because the i7 is not a processor made to do 3D graphics. The Cell processor wipes the floor with it easily, in 3D graphics and the other things I said. They simply don't compare and trying to argue otherwise is foolish. Likewise when it comes to general purpose computing and multitasking the i7 wins easily like I said before.

The Cell BE is-not-being-used-for-3D-rendering... due to many reasons...

Cell BE does not have access to VRAM. The GPU, however, can access VRAM and System RAM but does not have enough shading power. Cell BE can be used as additional shading units but it does not have access to VRAM... In this situation, the only solution would be to copy the rendered frame from VRAM back to System RAM so the Cell BE can work on it, then the GPU has to copy the rendered frame again from System RAM back to VRAM. VRAM usage is doubled in this case and even a portion of System RAM is occupied. The PS3 only has 256MB of VRAM so this operation is extremely costly, not to mention latencies between memory accesses and CPU <-> GPU operation.

I don't care what the Killzone 2 or any other PS3 developers have devised to work around this problem. Just ONE fact: the PS3 does not have enough RAM or memory controller access to allow the Cell BE to be a full-fledged 3D renderer. If anything, it can be used to transform vertices, sort z-order and alpha-test objects before sending final coordinates to the GPU to lighten a bit of the load. But again, that-is-not-3D-rendering.

Even Killzone 2 developers did not admit outright they used the Cell BE for 3D rendering. They only said that they used it to offload work from the GPU. The GPU can be used for transforming vertices, sorting z-order, and doing alpha tests as well.

Oh you want to play that game? Okay Consult a pcsx2 developer about how the i7 pales in comparison to the Cell processor for gaming. wait, no I did it for you already.

Sony's plans for PS2 and Dreamcast emulation on PS3

If you want to talk "gaming performance" then none of my above comparisons are relevant. When playing games natively (not emulated), 75-90% of the work is done by the Cell's 7 SPEs, depending on game. When emulating a PS2, those will be largely unutilized.

The SPE design is a very good one for gamedev in general. It'll murder the i7+GX290 without breaking a sweat, as far as gaming is concerned, assuming the programmers have a strong grasp of parallel programming concepts (which Sony coders certainly do). There are some other "flatter" multimedia tasks where the GX290 can hold its own against the Cell, but gamedev usually benefits nicely from the fact that SPEs are complete co-processors, and not just souped-up pixel shaders. (which is why the future generations of GPUs will most likely start moving in that direction)

This whole argument I could have just posted that, but I decided not to because I wanted to support my argument with my own words without copy & pasting what someone else said. But I guess this lays it all to rest. The Cell murders the i7 in gaming. Point blank period. End of discussion.

Did said developer work on the PS3 or is he just pulling marketting stuffs? I don't mean to offend anyone, but it doesn't look to me like he provided any data to back up his claims. When I mentioned PCSX2 developers, I meant those who actually provided data to their claims.

The AI is more complex than Left 4 dead believe that. It is not only about how the AI combats you but it is about how they react to you. If you shoot a guy in the arm, he'll actually react like he got hit in the arm, unlike in left 4 dead. In Left 4 dead, the zombies rush at you, you shoot them, they die and drop dead to the floor. GTA 4 not only has the elements of Killzone 2 where the people in it react realistically, they also have what Left 4 Dead has where there are lots and lots of people on screen.

The zombies don't just rush at you. They react to your footsteps, gunshots, and even random sounds like you kicking a can or you hitting the garbage bag. You can adjust all those parameters using the game console. It looks simple, but Valve did a lot of calculations in Left 4 Dead. Granted it wasn't the best implementation because the zombies always seem to just rush at you at whim, but they really did a lot of calculations. That's why the game runs at barely 30fps on the 360 and can drop down to 20 sometimes.

Steal a bus, and park them in Time square across the intersections to cause a traffic jam. Take out your AK 47, and then start shooting at the sidewalk full of people, they will all start running away from you, all of the cars will start to go out of control scrambling to get away from the carnage while you are shooting the sidewalk full of pedestrians. One of the pedestrians you hit got hit in the arm. He starts holding his arm in pain. Then he looks at where you shot him, then he looks straight at you in utter fear with a terrified look on his face. Another guy gets hit from the bullets while he was running, he gets hit in the leg. He falls over and hits the ground realistically. then he struggles to get up and he starts limping away in fear. The police come and they all are trying their hardest to shoot you down to save the people who haven't been killed yet. All of this happening all at once. Lot's of people on screen, lots of things happening at once. And in general it is just a very very horrific and sad columbine style massacre.

Half of that is already done in GTA SA. Only the emotion parts are not done. Guess how GTA SA runs.

But granted, again, I tried GTA IV again, and after the patch, it got better. But here's the thing: it still runs slower than it should.

They specifically said that it was too complicated and that they didn't want to put in the effort. I don't think that they said that Left 4 dead can't be done on the ps3. Can you post a link to them saying that it can't be done?

Fact is the game doesn't exist on the PS3. Done. I apologize and assume it's not impossible if they take out all of the extra processing they put into the game, BUT... the game as it is right now is so hard to port that Valve themselves didn't even bother. But if they said it was impossible for them then people would start bringing up the Orange Box...

Even if they said that it can't be done, I don't think that they would be an authority on what the ps3 can and can't do since they actually never made a game for the ps3. The Orange Box on the ps3 was was made by a different company according to what they said.

If they wouldn't be then take Id's words on it.

By that logic GTA 4 is gpu limited because I run it at 2560 x 1600 and moving from a 9600GT to a GTX 280 greatly improves my performance.

Then it's GPU-limited.

But only where streaming is not concerned. The game still streams an insane amount of data over using the CPU so even if it is indeed GPU-limited (well, it should be), the streaming should not be so bad. But it is... sadly. This game would have worked out very well if they would just optimize the streaming somewhat. About 50% of its slow-downs now are due to streaming... from what I can see after further testing. But that, again, means that it's unoptimized. GTA SA could be streamed so flawlessly.

It's one SIMD unit. With 128 registers per hardware thread. The 360 has 6 hardware threads so in all there are 768 registers. Search google. Matter a fact here.

Xbox 360 128 registers per hardware thread - Google Search

I did it for you. You don't get just one website but many websites.

Okay, I'm wrong. And I apologize for that.

If you look at the Killzone 2 debug info, it shows you that the cpu is barely stalled by the spus. The PPE is good enough to take on the task.

Because the SPUs are not utilized much. But the game is not open-world, does not have a lot of things done even to its physics. It has ragdolls, yeah... but not everything in the game can be blown off. Frame rate dips a lot sometimes, and enemy AIs are very predictable.

I played it... and I know how it is...

No, they are typically slower because the 360 is easier to program for and it is easier to get acceptable performance from it.

Well, more like people can't get acceptable performance out of the PS3 without compromising stuffs.

Crysis 2 runs just fine on the 360 and ps3 without having objects sink into the floor.

Uhh... actually, Crysis 2 performance on 360 and PS3 currently are below acceptable.

Nope. It doesn't. Show me an example of a pc gaming running faster on the pc with an i7 and a 7800GTX 256MB.

Street Fighter IV
Devil May Cry 4
Lost Planet

No, seriously. Try. Those games run better than on a PS3 even when they are coupled with just a Core 2 Duo.

Why don't you just prove your point. i'm not going to get 2 versions of Street fighter 4 and a 7600GT to prove the point you made.

7800GTX. Not 7600GT.

Granted, the GPU in the PS3 is tweaked and has a higher core clock even, but I bet you'll find to your dismay that the game actually runs faster on PC.

Or actually, run GTA IV to test even. After the patch, the game got a a bit better, though it should still run faster than now... I think you can run it faster than on the PS3 or 360 at least.

[Mr. Duck Sauce]

bkwegoharder just fails on what he is talking about since its more opinion based than actual data or understanding about how things are run, please use more rational facts rather than fanboy opinion.

__________________
AMD Athlon 64 X2 7750 2.7Ghz Kuma
2Gb DDR2 800 Mhz
XFX 9800GTX+
WinXP 3

[Dennemark]

I take the Phenom cuz i hate intel's pricing :P

__________________

Originally Posted by Kirby

one time I wasted my master ball on one

then I let mewtwo get away....I was suck a dumbass

Take the Magic: The Gathering 'What Color Are You?' Quiz.

[Yogi]

Finally, gotten my Phenom II x3.

[KrossX]

Congratulations.

__________________