PCSX2 Review Part2 : GPU Power Also Matter

[RPGW1ZaRD]

PCSX2 Review Part2 : GPU Power Also Matter

Some of you maybe remember my previous PCSX2 review: CPU power DOES matter where I compared dual core performance with single and different clock speeds. But how about graphic cards and PCSX2? I've heard it so many times, graphic card doesn't matter in PCSX2 only the CPU does like often is the case for emulators. But is it really true? Today we're looking at a few bit more demanding games graphics wise and see how my previous overclocked 6800GT 256MB will compare to an overclocked slightly past 7900GTX performance 7900GTO 512MB card. Do we see a difference between even these relatively powerful cards, I think it's natural to say also the GPU performance will impact on PCSX2 frame rate. The reason I got interested in this started with a couple of threads conserning about low performance despite having a highend dual core CPU and a couple of questionmarks I've had in my head regarding the CPU usage. I thought it's weird how Core 2 Duo CPUs clocked past 3GHz didn't beat my Opteron 165 @ 2.8GHz even tho they should and despite people have tried everything to get better speeds, then I noticed the users in question were only using graphics cards like 7600GT or similiar and their CPU usage were often low, so that left me one questionmark to be observed, how much and if graphic card performance matter. I got bored of all the rules-breaking threads as of recently so I felt I need to be creative and what I deal with here is as familiar for me as my own room.

OK, enough of the babbling, grab your favorite reading glasses and some popcorn and let's get started:

Test setup

CPU : AMD Opteron 165 1.8GHz @ 2.8GHz 2x1MB L2 cache
Motherboard : DFI Lanparty nForce4 SLI-DR
GPU1 : NVIDIA 6800GT 256MB PCI-E 350/1000MHz @ 440/1155MHz
GPU2 : NVIDIA 7900GTO 512MB PCI-E 650/1320MHz @ 700/1520MHz
RAM : OCZ Gold VX PC4000 2x512MB @ 255MHz 2-2-2-5-1T (settings)
PSU : OCZ PowerStream 520W
Sound : Creative SoundBlaster Audigy
OS : Windows XP Pro SP2; AMD Opteron driver, MS hotfix
and dual core optimizer installed
Cooling : Big Typhoon 120mm CPU aircooler and 9x case fans

PCSX2 config

PCSX2 Build : v0.9.1 VM build
CPU : EERec, VU1rec, VU0rec, MTGS, Dual Core Mode enabled
GFX1 : ZeroGS KOSMOS 0.95.2; 1024x768, FFX Hack (only in FFX/-2),
AA4x & Bilinear filtering (both low quality and high quality shots included)
GFX2 : GSdx9 SSE2 0.9.0; windowed, ps1.4 (unless stated otherwise),
LTF, NLOOP 0 hack (only in FFX/-2)
Sound : P.E.Op.S. SPU2 DSound 1.4.0; fastest settings
CD/DVD ROM : Linuzappz Iso CDVD 0.5.0

Test method

This time I want to be more precise so I included FRAPS FPS value that shows more like an average FPS value of the frames rendered every 3-5 seconds or something like that and I will use this value in the graphs. I also try to take note about the polygon count much more and I can promise an error margin of probably within 1 FPS. I take screenshots from different areas and compare them in different ways with the 2 graphic cards. Games used are WWE SmackDown vs. Raw 2006, Resident Evil 4, Tekken 4, Grandia III, Final Fantasy X and Final Fantasy X-2. The first 4 games meantioned are much more graphics heavy games than the last 2 in order to at least show a little variation in the results and to approve my conclusion about this test. I also made the review more 56k-friendly this time so I will provide links to the screenshots instead for those interested checking them out.

Test1 : WWE SmackDown vs. Raw 2006

Screenshots (Sample)

This is the game that finally actually revealed to me that GPU power also can matter significantly, it all started with this thread for those interested. Note especially his screenshot on the 1st page and my own on the 2nd page of roughly same spot. This would be a very useful spot for this review if it wasn't a scene in motion since it's hard to get any very exact comparisions. Even my overclocked 7900GTO card is bottleneck here. Anyways for those who are lazy to check out the thread his specs are Core 2 Duo E6400, 1GB DDR2 and NVIDIA 6600 and check these pics: his rig: 6600, my rig: 6800GT, my rig: 7900GTO. You had expected a tad more of his CPU right? Well the answer is the graphics card here, isn't it a lovely difference between the cards, ~128% performance advantage for 6800GT over 6600 and ~162%! difference between 6800GT and 7900GTO or as much as a whooping ~498%!! difference between 6600 and 7900GTO! Believe me I double checked so everything was like normal and only thing that was different when I did that comparision between 6800GT and 7900GTO was the graphics card. I really had a hard time believing what I saw then tho but at same time it opened my eyes of that PCSX2 uses much more GPU than I had previously thought.

Now let's move on, the spot I use here in this review won't show the same dramastic difference since in this spot the 7900GTO is powerful enough so the CPU becomes the bottleneck, thus the CPU usage went from 70% with the 6800GT card to 100% using 7900GTO. Even if it wasn't a whooping 162% difference like in the other spot due to my CPU becoming the limiting factor, it's still a nice 87,5%! advantage for 7900GTO which can be compared with like an overclock of Opteron 165 1.8GHz to ~3.4GHz, in other words a remarkable difference. I wouldn't bother running this game with AA4x and bilinear filtering with today's graphic cards though as the 7900GTO is only able to maintain 17 FPS vs 6 FPS for 6800GT, but it is a nice ~183% difference between the cards though and only ~41% behind 6800GT in low quality mode.

Test2 : Tekken 4

Screenshots

Screenshot 1 (Sample)

Tekken 4 is a good game to use in this test apart from having a few graphical glitches and can be run in both GSdx9 and ZeroGS with MTGS & Dual core mode.

The jungle arena for example is very demanding gfx wise and it's stressful for any computer today. Since this game seems to run better with palletized textures which requires pixelshader 2.0+ support I wouldn't bother testing this game if I hadn't at least a decent fully supporting pixelshader 2.0 card. Palletized texture setting seems to significantly lighten the load on the graphics card as CPU usage climbs a lot. Running it with pixelshader 1.4 and palletized disabled like usually gives best speed in GSdx9, in this arena it crawls at just above 2 FPS even on this computer, so palletized textures + ps2.0 offers an oustanding 7.5x or 650%! better FPS rate for 6800GT and 11.5x or 1050%!! better FPS rate for 7900GTO. Between the cards 23 vs 15 fps is a not too shabby ~53% difference in this mode and 64% comparing ZeroGS low quality mode and in AA4x+BF it's a bigger 117% difference.

Screenshot 2 (Sample)

This spot is also quite interesting and it shows various areas in games can be much less demanding as it's the fastest running spot in the game and the only arena where graphics isn't that heavy and CPU becomes the bottleneck and that's why CPU usage is 100% and the FPS is a bit higher here even for 6800GT. Both cards comes head and head in GSdx9 but with ZeroGS there's a ~14% advantage for 7900GTO. 6800GT doesn't quite cut AA4x+BF as usual though coming behind 7900GTO 8 vs 18 FPS with ie. 125% and CPU usage has also dropped majorly showing the graphics card is a huge bottleneck.

Test3 : Resident Evil 4

Screenshots (Sample)

Resident Evil 4 is another graphics demanding game which will require both a powerful enough CPU but also graphics card in order to run at playable speeds. Once again you see in the graph that GSdx9 is a less graphics card dependant plugin, where 6800GT can actually manage a quite good job in GSdx9 being only about ~16% slower than 7900GTO and CPU usage isn't that much lower either. However with ZeroGS plugin the difference becomes bigger where 7900GTO is 50% faster than 6800GT in low quality and here the CPU usage varies also more between the 2 cards, showing that ZeroGS is a more GPU dependant plugin. AA4x+BF becomes just too much for both cards as CPU usage only is at 17%! with 6800GT and 47% with 7900GTO and 7900GTO has a ~83% performance advantage here too.

Some of you might remember a picture from the Screenshot Contest with a picture from roughly same area with a Conroe clocked at 3.6GHz and only had about 22 FPS in that screenshot using GSdx9. Now you're asking why do I have even higher than that then with my CPU which is slower than his and the answer is the graphics card once again, he only had a relatively slow graphics card (can't remember exactly what he had) which is the bottleneck. I can assure you that even lowering my CPU overclock a bit won't affect the FPS rate in this game among other very demanding games as the GPU is the limiting factor. Also at same time it doesn't help overclocking the CPU in that case either, you can even have a 4GHz Conroe and still get same FPS rate if the graphics card is not fast enough. Overclocking the graphics card in games like this can help a bit tho, not only when enabling AA4x+BF, check: 7900GTO stock vs 7900GTO overclocked which in this case is a ~12.5% performance difference between stock and overclocked speed.

Test4 : Grandia III

Screenshots

Screenshot 1 (Sample)

Grandia III isn't quite as demanding as the previous games in this review but still belong to the a bit more GPU demanding games. In this first screenshot 6800GT is the bottleneck and with the 7900GTO, Opteron 165 @ 2.8GHz becomes the bottleneck with 81% vs 100% respectively CPU usage and the difference between these cards is ~48% which would be much bigger difference if I had even more powerful CPU which would cause the CPU usage to stay below 100% and show the 7900GTO is still the bottleneck. AA4x+BF mode shows an interesting result here too where 7900GTO is ~211%!! faster than 6800GT.

Screenshot 2 (Sample)

This spot I wanted to include as it reveals a similiar pattern in the results as the previous screenshot. CPU usage and FPS rate is nearly same as in the other spot which more precisely proves how the CPU usage relates to GPU performance. 6800GT is ~37% slower than 7900GTO and CPU usage is very similiar like in the first screenshot 82% vs 100% respectively. Here there's an amazing 250%!! performance difference in AA4x+BF mode though (perhaps the 6800GT thought the fog was too much and reminded it of smoke as in a smoking warm GPU which it is literally speaking as I've had it at 98C once with stock cooler without any complaints from the card just so you know :laugh:). Right, let's move on to:

Test5 : Final Fantasy X

Screenshots (Sample)

Final Fantasy X was included since it's much less GPU dependant game just to show the graphics card doesn't matter that much in every game, well at least if we're speaking about GSdx9 and ZeroGS in low quality mode, in AA4x+BF mode you'll still need a decent card not to get huge FPS drops.

The graph shows that in both cases I am CPU limited in ZeroGS in low quality mode as both FPS rate and CPU usage is the same. But when we check AA4x+BF mode there was pretty huge 102 vs 41 FPS or ~149% difference, where 7900GTO is almost able to handle it without any FPS drop, however overclocking the card further to what I run the card all the time at now pretty much almost fixes this though, check: 7900GTO stock 650/1320MHz, 7900GTO 700/1520MHz (=speed used in this test), 7900GTO 690/1650MHz (=what I run this card at all the time now), so between stock and the speed I daily use there's a ~18% performance difference and CPU usage got very close to constant 100% so the GPU bottleneck almost disappeared.

Final Fantasy X is not that much of a GPU dependant game however, I'm able to run maybe 60 - 70% of the time with AA4x+BF without any slowdowns with the 7900GTO card. Most often the slowdowns appear in battles but in some battles I can enable AA4x+BF while in others I need to use AA2x for example and in some I need to disable both bilinear filtering and antialaising in order to not be GPU bottlenecked. 6800GT is too slow in any area in order to AA4x+BF mode not slowing down at all.

Test6 : Final Fantasy X-2

Screenshots

Screenshot 1 (Sample)

Final Fantasy X-2 is a bit more GPU dependant than what Final Fantasy X is but still not close to the other games used in this test. In most areas I have to disable bilinear filtering while I can leave AA at 4x though without getting any FPS drops other than in some battles.

In the first spot we see both cards are enough for this spot in low quality mode as CPU usage is at 100%, but when we check AA4x+BF mode CPU usage drops to 63% with 6800GT and 76% with 7900GTO while performance difference is a huge 155%!

Screenshot 2 (Sample)

This spot is one of the most demanding spots if not the most demanding I know of in X-2 and here we also see a slight difference between the cards in low quality mode, 28 vs 35 FPS or a ~25% FPS rate difference and neither of the cards were enough for this spot in order for CPU to become the entire bottleneck despite low quality mode. In AA4x+BF mode the CPU usage drops further and this time it's a remarkable ~93% difference.

Conclusion

What I wanted to show with this review is that GPU power also can matter in PCSX2 quite remarkably. Of course this doesn't mean ALL games will require a powerful graphics card, Disgaea for example might even run good enough with a Nvidia FX serie or Radeon 9500+ card. It's just that some games are much more demanding than others. Some seems to require even a 7900GTX performing card in order to provide playable speeds and for some even an Nvidia FX5500 might be just enough. Before I used to think maybe a Nvidia 6600GT card would be enough for all games in ZeroGS low quality mode or GSdx9 but seems I can finally throw that idea into the grave. One thing remains the same as always though, if you've got a slow computer you're far better off playing on your real PS2 unless you want to spend lots of money just to be able to run at somewhat enjoyable speeds.

Think of the CPU and graphics card like they would be both some people carrying a heavy piano for example, the one of them that isn't as strong will put the limit at which rate they are moving the piano. It doesn't matter if you had a strong person paired with a weaker one as the strong person is dependant on the other person and can't move it himself without him. This is why it's stupid to pair a weak GPU with a powerful CPU for example no matter if we talk about PCSX2 or PC games, well balanced parts are always the best and gives you most value for the money.

What you also learn by this review is that you can check whetever you're CPU limited or GPU limited in the games (taken into account your system is running like it should) by checking CPU usage in windows task manager. If CPU usage is at 100% it means you're CPU limited and if it's below 100% it means you're bottlenecked by the graphics card (EDIT: Seems this only applicates to PCSX2 v0.9.1, can't garantuee future versions behave the same way).

I hope you enjoyed this review as I spent quite a lot time with it but it was interesting and worth it. Who knows maybe I'll make a 3rd part sometimes talking about how RAM timings impact PCSX2 performance but for now I just sit back for a while and perhaps test out some more PS2 games on the lovely PCSX2 Playstation2 emulator. :thumb:

Links to the package containing all material used in this review:

Link1
Link2

[Alan_Brito]

One more time an excellent contribution to the forum. Thanks man!!!

[KnutSJ]

Great work! Two thumbs up. (I'd gladly give more thumbs but apparently I only have two)

[Haru]

nice
thx RPGW1ZaRD for this review

I'm waiting for my ASUS P5W DH DELUXE/WIFI + AS5.
When they arrive im going to push my CPU a bit further ;D

[cyclonmaster]

Nice review

[Qwin]

Great review indeed very nice

[smokemybong]

excellent review must have taken you awhile though good job!

[Silenus]

Good Job on this.
Didn't know you were |Gand|Alf| before, until now.

[Barius]

Were you planning on buying an 8800? I'd love to see the results for that card

[refraction]

Both reviews are now linked in the FAQ good work!

[xJOSx]

well, at least I know I can get some more fps later on... Maybe I'll upgrade my 6800 GeForce Go Ultra to the 7950 or 8800 when prices drop more... thx for the news!

[RPGW1ZaRD]

Thanks

Barius> Sorry but my budget doesn't quite allow me every 6 months upgrades yet as I'm only a student (altho on holiday atm ) so I'll probably wait for Nvidia 8900 series card which should be released at least by mid 2007 but perhaps somewhat sooner. Then it's prolly time for a major upgrade hopefully to an AMD K8L quad core system perhaps even, should be sweet 8900 + K8L.

8800 at launch is ridiculously overpriced imo too, better wait a while and you'll see both GTX and GTS pricedrops of probably up to $100 or so.

[Fire]

good work

[Vlado]

Greate job , i finally realised what a sucky card my 5200 is ( cheers , and thanks for the hard work !

[Traxxx]

Again a great and detailed testing by you

[DeserTwolF]

excellent work!

[laynlow]

Thanks for taking the time to do that great review

[ulwan25]

nice reviews! maybe DX10 will boost up little FPS

[El_Diablos]

RPGW1zarD thx for nice review! Cool to know

[RPGW1ZaRD]

Originally Posted by ulwan25

nice reviews! maybe DX10 will boost up little FPS

I do believe with all the new features and improvements in DirectX10 could really benefit PCSX2 a lot but the DX10 graphic cards are already powerful to maintain good performance in DX9.

But the main problem here is the DX10 adapting, I don't expect every1 to switch to DX10 cards immediatly and neither do I think it would be that easy for PCSX2 coders to adept to the DirectX10 features and how it works that soon and could be somewhat time consuming. But the way it works and the features offered could very well benefit PCSX2 a lot as the efficiency seems to have been improved a lot in DX10. But since the DX10 cards already comes with lotsa raw power and DX9 cards couldn't benefit or use the features the meaningness of it becomes much less. This would of course lead to that plugins and emulator to be rewritten to be able to benefit from it so it's nothing if I was a coder that I would consider be any important area to focus on for now but later on when time is right I'd look into it.