Scoob
New member
Hi All,
I was a little bored this evening so fired up Just Cause II for a quick random blast. Nothing unusual there but I'd been playing earlier with other things so had BOTH my screens enabled - the second screen showing Task Manager, CPU load, Core Temp, GPU-z (2x, 570 SLI) as well as CPU-z. Basically my usual monitoring stuff.
Anyway, I thought I'd fire up the "Dark Tower" benchmark just to test my system as I'd not run JC2 for quite some time - and not on my new W8 build. I noticed some interesting things...
It seems that any game that uses PhysX cannot spread the load over multiple GPU’s. I observed in the Just Cause 2 benchmark that one GPU was 20% more loaded than the other. I traced this to the “GPU Water Simulation” setting that appears to use PhysX. Turning that setting OFF saw perfectly even GPU load in the benchmark – so the CPU handling the water simulation.
On the “Dark Tower” benchmark, I got 120.1 fps average with the CPU handing the water simulation (so "GPU Water Simulation" set to OFF) – not a bad score as JC2 is a fairly demanding title with all things turned up.
Running “Dark Tower” again, but this time with GPU Water simulation enabled saw the uneven load once again, so one GPU in the high 90’s consistently the other in the high 70’s consistently. This gave me an average FPS of 77.19 – a huge drop!
So, PhysX processing does NOT scale over two GPU’s, it just uses one. Yes, my settings are all Auto for this, I've not selected one GPU. In SLI that means the non-physx card is in effect nerfed by how much load the card allocated to PhysX - and general rendering of course - uses. Basically, if in game both cards sit at say 90% constantly, the extra PhysX load on ONE card would add 20% (as it did for me) meaning one card would be at 90% WITH PhysX work being done, the other would “lose” 20% and be stuck at 70% – hammering the FPS. In effect both cards are rendering at 70% GPU load, whereas the remainder on ONE GPU is hadning the PhysX.
I suspect that as my CPU is still pretty good, and of course overclocked to boot (2500k @ 4.6), I ironically get WORSE FPS with PhysX enabled as only one of my cards is utilised. This suggests that potentially a THIRD GPU for dedicated PhysX would do the business, but that's a bit silly lol.
To test things further I thought I'd fire up Mafia II as this has PhysX options - set to HIGH – as well as a Benchmark mode of course. Already I can see the uneven GPU load, just 15% difference here though, so not quite as severe as in JC2.
So, WITH PhysX on High – average FPS is a rather crap 55.8 – I have scored MUCH better than that in the past. This gives me a Rank C – I usually get a Rank A!
So, testing again but with PhysX DISABLED – average FPS is an acceptable 104.3, giving my system a “Rank A” rating. GPU load was generally pretty even, though there was a marked increase in CPU load.
I'm a bit of a nerd, so I found this pretty interesting. There are few titles that make use of PhysX, well, not that I own at least. I've always enabled it in games when it's available as I assumed it was giving me a boost, an NV GPU being much better at this than a CPU. However, as this does not appear to scale over GPU's at all, and those of us who run SLI likely have a fairly good CPU, PhysX appears to be detrimental to overall performance.
If you have a SINGLE GPU and say a fairly low-end or maybe just not overclocked CPU, then I can see how PhysX might give a boost. To the enthusiast running SLI however PhysX seems rather pointless - at least from my observation of these two basic tests.
I think what has happened is that CPU's took a bit of a quantum leap forward in terms of performance a little while ago. I remember being perfectly happy with my old overclocked AMD 4200x2 - it ran everything I wanted & I was quite content with the performance. However when, in a moment of spending madness (sanity?) I got myself a Q6600 and 8800GT based system I was BLOWN AWAY by the improvement. I remember even taking my new build down to a friend, who had a very similar system, and he started ordering his new PC bits there and then! My Q6600 system did me proud for years, it overclocked well (on air to 3.6) and the addition of a GTX 275 and finally a GTX 570 improved its longevity no end. It's now the PC I'm typing this on btw, working as my general Gateway / Server PC.
When I finally did move on, in another "I'm bored, need STUFF" moment to my 2500k system I was blown away once again by the improvement. As I was, once again, after getting a 2nd GTX 570 - I even posted here about my first SLI experience I was so pleased.
Anyway, enough rambling from me, in essence I'd say that the progress in higher-end CPU's has largely nullified the advantages of PhysX in high-end system. This makes me wonder if how things will be with Direct Compute moving forward. If DC scales nicely over multiple GPU's then maybe us SLI / Crossfire users will reap the benefits. If, however DC uses a significant enough chunk of extra computing power - power that our modern, fast CPU's could easily provide, maybe we'll be worse off. Note that I am of course talking purely gaming here, my understanding is that Compute performance is critical in things such as Folding at Home and this is a GOOD thing.
So, my ramble really is over now, as always I'm interested in others thoughts
Cheers,
Scoob.
I was a little bored this evening so fired up Just Cause II for a quick random blast. Nothing unusual there but I'd been playing earlier with other things so had BOTH my screens enabled - the second screen showing Task Manager, CPU load, Core Temp, GPU-z (2x, 570 SLI) as well as CPU-z. Basically my usual monitoring stuff.
Anyway, I thought I'd fire up the "Dark Tower" benchmark just to test my system as I'd not run JC2 for quite some time - and not on my new W8 build. I noticed some interesting things...
It seems that any game that uses PhysX cannot spread the load over multiple GPU’s. I observed in the Just Cause 2 benchmark that one GPU was 20% more loaded than the other. I traced this to the “GPU Water Simulation” setting that appears to use PhysX. Turning that setting OFF saw perfectly even GPU load in the benchmark – so the CPU handling the water simulation.
On the “Dark Tower” benchmark, I got 120.1 fps average with the CPU handing the water simulation (so "GPU Water Simulation" set to OFF) – not a bad score as JC2 is a fairly demanding title with all things turned up.
Running “Dark Tower” again, but this time with GPU Water simulation enabled saw the uneven load once again, so one GPU in the high 90’s consistently the other in the high 70’s consistently. This gave me an average FPS of 77.19 – a huge drop!
So, PhysX processing does NOT scale over two GPU’s, it just uses one. Yes, my settings are all Auto for this, I've not selected one GPU. In SLI that means the non-physx card is in effect nerfed by how much load the card allocated to PhysX - and general rendering of course - uses. Basically, if in game both cards sit at say 90% constantly, the extra PhysX load on ONE card would add 20% (as it did for me) meaning one card would be at 90% WITH PhysX work being done, the other would “lose” 20% and be stuck at 70% – hammering the FPS. In effect both cards are rendering at 70% GPU load, whereas the remainder on ONE GPU is hadning the PhysX.
I suspect that as my CPU is still pretty good, and of course overclocked to boot (2500k @ 4.6), I ironically get WORSE FPS with PhysX enabled as only one of my cards is utilised. This suggests that potentially a THIRD GPU for dedicated PhysX would do the business, but that's a bit silly lol.
To test things further I thought I'd fire up Mafia II as this has PhysX options - set to HIGH – as well as a Benchmark mode of course. Already I can see the uneven GPU load, just 15% difference here though, so not quite as severe as in JC2.
So, WITH PhysX on High – average FPS is a rather crap 55.8 – I have scored MUCH better than that in the past. This gives me a Rank C – I usually get a Rank A!
So, testing again but with PhysX DISABLED – average FPS is an acceptable 104.3, giving my system a “Rank A” rating. GPU load was generally pretty even, though there was a marked increase in CPU load.
I'm a bit of a nerd, so I found this pretty interesting. There are few titles that make use of PhysX, well, not that I own at least. I've always enabled it in games when it's available as I assumed it was giving me a boost, an NV GPU being much better at this than a CPU. However, as this does not appear to scale over GPU's at all, and those of us who run SLI likely have a fairly good CPU, PhysX appears to be detrimental to overall performance.
If you have a SINGLE GPU and say a fairly low-end or maybe just not overclocked CPU, then I can see how PhysX might give a boost. To the enthusiast running SLI however PhysX seems rather pointless - at least from my observation of these two basic tests.
I think what has happened is that CPU's took a bit of a quantum leap forward in terms of performance a little while ago. I remember being perfectly happy with my old overclocked AMD 4200x2 - it ran everything I wanted & I was quite content with the performance. However when, in a moment of spending madness (sanity?) I got myself a Q6600 and 8800GT based system I was BLOWN AWAY by the improvement. I remember even taking my new build down to a friend, who had a very similar system, and he started ordering his new PC bits there and then! My Q6600 system did me proud for years, it overclocked well (on air to 3.6) and the addition of a GTX 275 and finally a GTX 570 improved its longevity no end. It's now the PC I'm typing this on btw, working as my general Gateway / Server PC.
When I finally did move on, in another "I'm bored, need STUFF" moment to my 2500k system I was blown away once again by the improvement. As I was, once again, after getting a 2nd GTX 570 - I even posted here about my first SLI experience I was so pleased.
Anyway, enough rambling from me, in essence I'd say that the progress in higher-end CPU's has largely nullified the advantages of PhysX in high-end system. This makes me wonder if how things will be with Direct Compute moving forward. If DC scales nicely over multiple GPU's then maybe us SLI / Crossfire users will reap the benefits. If, however DC uses a significant enough chunk of extra computing power - power that our modern, fast CPU's could easily provide, maybe we'll be worse off. Note that I am of course talking purely gaming here, my understanding is that Compute performance is critical in things such as Folding at Home and this is a GOOD thing.
So, my ramble really is over now, as always I'm interested in others thoughts
Cheers,
Scoob.
