The Somewhat Definitive "Are 3+ cores better than 2" PCSX2 thread.
#1
First off, I know PCSX2's frame counter isn't the most accurate, but it should be accurate enough to prove a point.

Several members of the forum have postulated or thought they had seen improvements when using more than 2 cores for pcsx2. In theory having an extra core or two should improve speed if only because OS and anything in the background should be running on 1 thread, with PCSX2 running on 2 free ones... But in practice does this actually amount to anything? And if so how much?

I chose 3 games from my collection at random with frame limit turned off. The computer system is as follows:

Phenom X3 720 BE @ 3.5ghz for this test
2GB 5-5-5-12 DDR800@920
ATi Radeon HD3850

First game, Final Fantasy 12.

First off is the dual core test... 97fps quite impressive

Triple core gets a noticable boost, about 10%

Second game, Mega Man X: Command Mission a scene from a battle early in the game.

Dual core: 130fps Triple core: 135

This was much less noticable, so much so it is nearly statistically equal.

Final test, Persona 4..

Dual core: 98fps Triple core: 110fps

Once again about a 10% difference.


What's this mean to the average person? If you're looking for a new processor with PCSX2 being the main or only thing on your mind a higher clocked dual core (which is what everyone in general already knew). If you're trying to decide between a dual and a triple/quad and the latter isn't too much lower clocked than a dual it should perform just as well.


Stay tuned as I also plan on dropping the thing to 1 core just for giggles... and if there is any demand, I'll try and play with clockrates and see at what point a dual core matches speed more or less 1:1 with a higher multicore.

(Edit) Single core tests for giggles. MTGS is still enabled (IE they still use 2 threads) but I turned off 2 of my cores in the bios.

Pictures included, all games report about a 50% faster speed dual-core versus single core which is completely expected. Really surprised all 3 test games run about the same speed as my old AMD x2 5000 in dual core mode. Neat.

(edit 2) Final conclusion: If you're going for a system specifically for PCSX2 and you're on a budget, you won't get enough of a benefit from going tri or quad. Higher clockrate is much much more important than number of cores. That said, if you're building for a general computer and you want some "future proofing" (I always find that phrase hilarious when it comes to computers) you can't go wrong with a decent clocked quad (or at least one that's easy to overclock ^_^)


Attached Files Thumbnail(s)
                       
           
[Image: 2748844.png]
Reply

Sponsored links

#2
Interesting, but do you know how the processor's cache behaves with more/less cores? It could be just cache usage changing, causing the FPS differences.
Tips: Choose totally static places to keep your fps from fluctuations, repeat each test about 3 times and take the average of the FPS.
Also check native in GSdx to have the GPU do as less work as possible so it won't interfere with your results.
Oh and do use FRAPS for the FPS, you can use it in free mode, should be way more accurate
[Image: newsig.jpg]
Reply
#3
All three scenes were static. The phenom's have 512k L2 per core, and 6mb of shared for all cores (1, 2, 3, or 4 depending on what's running) so if it was only a matter of CPU cache, 2 cores would in theory be faster as each core would have 3mb of cache versus 2mb for tri cores.


While the FPS isn't perfectly accurate, it still can demonstrate that there is a palpable difference. I'll redo the tests real quick with fraps just to get a confirmation on the results.
[Image: 2748844.png]
Reply
#4
10% in 2/3 games and 0% in 1 isn't telling much really, it's in the margin of error Wink
Don't forget 10% is at 98fps vs 110fps. Try testing a very cpu intensive game that gets you say 40 fps max and see what the difference will be then (probably 0)
It could be affecting for example only a 50% of games (or lower) but we really can't know for sure. It is also dependent on what you are running at the time (I realize you did the test with the same running programs in all cases) but maybe for a user who has lots of progs running it would make more difference or vice versa.
Pretty hard to accurately measure the effect of the extra core, but from a programmer's point of view it should be none to very small (pretty close to your results)
[Image: newsig.jpg]
Reply
#5
True true, I definitely wasn't trying to sell people on the fact that any extra cores will do a lot of good, quite the opposite.

-laughs- Guess I really shouldn't call the thread definitive eh?

-edit- Also how inaccurate would you say the FPS meter is? As fraps is showing nearly identical FPS to what the meter shows, albeit maybe not perfectly in-sync (generally delayed maybe a tenth to a third of a second)


Attached Files Thumbnail(s)
   
[Image: 2748844.png]
Reply
#6
Hehe well do more tests, we are all learning anyway. There are so many parameters that could affect stuff that you can practically never know unless you give it a try Wink
[Image: newsig.jpg]
Reply
#7
One benchmark I've been personally trying is with GT4.
I create a new directory for Pcsx2 with all the plugins I want to test, then using pcsx2ConfigSaver I configure the plugins in 3 ways. The first is with no hacks on, GSdx in native res, DX10 Hardware, and nothing else, to test the CPU. The second is the same as the first but with GSdx using 3x internal res of GT4, this is more to test my GPU. And the last is exactly how I normally use the emulator, using hacks and various GSdx settings, but with no CPU limits, and is more for me to test if my PC is capable of playable fps in the games I play.

I do a lap of a circuit like the New York City circuit, save the lap to a new memcard then do a save state at a point where I can quickly do a replay of my lap, noting the exact time it takes to do a lap. Next I run fraps, put in the amount of seconds it takes to do a lap in the "Stop Benchmark automatically after" box, tick the MinMaxAvg and FPS boxes, assign a benchmark hotkey, and run my test hitting the key just as you start a lap.
It's nothing like having a real benchmarking program, and theres a little messing around, but it's quite surprising how different the end results are.
One thing I've found in testing with GT4 is that, from Pcsx2 builds after 1732, Anisotropic Filtering is broken, or appears to be broken.
With build 1732, I can use many different builds of GSdx and AF works when I have it forced in the CP, to 16x for example. But after 1732, any build so far, and any GSdx, it's broken. Is this a known thing ?
Just wondering.
Intel E7500 @ 4.00ghz 400 fsb / Asus P5QL Pro / 4Gb Kingston RAM / PNY nvidia 9800GT 512Mb / Creative X-Fi Music 24 / Vista 64 SP2/
Reply
#8
If you can be bothered, you can try experimenting with CPU affinities. You could either use Task Manager or another program to change the affinities of all other processes to one core, and change PCSX2's affinity to the remaining 2 cores. Don't know if that'll produce better results or not.
[Image: yunacopy.jpg]
Reply
#9
I was wondering about things like this too. On my i7, I limited the affinity of fight Night Round 4 to 2 cores, then I put the affinity to all of my cores, and the fps went up. It was by like 2 fps, but it was easy to tell since whenever I turned the afinity back down to 2 cores, it would immediately drop back down by 2 fps, then letting it run on all cores would immediately have the fps rise back up by 2. This was in a pause screen.
Reply
#10
i just had the same idea to play with affinity and priority.
I used MGS2 and the scene was pretty static.
i have Phenom II 940, win7
2 cores - 50% of speed
3 cores - 77% of full speed (sound stops being cracky)
4 cores - 77% of full speed (almost no speed gain)
increasing priority seems to give 1-2% of speed (could be a fluke)
Phenom II X4 940 3 Ghz / 8 gb RAM 800 / Geforce GTX 460 / win7 64
Reply




Users browsing this thread: 1 Guest(s)