A question I could not answer, nvidia + physx.

With a guess I would say it would effect performance, because now the cards instead of just rendering are now computing physics also. The orginal Ageia cards were Pci-e 1x slots if i remeber right so its not allot of bandwidth but still extra workload.

If what you're getting at is as follows:

If using the Physx option during SLI will drop the usage of the combined GPU performance then the answer is yes and no.

Using SLI allows you to share the load and it counts in most cases a lot when using high res. By adding the Physx portion you will get a hit. The question is to what magnitude? If you're running a game with a stable FPS of 60 and taking a 10fps hit then it really doesn't matter. If you are doing 30 and take a hit of 10 then it's more profound. Note I'm using examples here though.

As for how much it takes off the combined gpu performance only he will be able to tell. Note that doing physx calcs on a GPU vs a CPU is quite different - the GPU threaded architecture allows for physc calcs more easily than a CPU does as a CPU still has added work to do. So doing such calcs on the GPU can happen very fast through a small section of the streams. And even then it's open to debate to what extent the calcs are done based on the game (how much physx is done).

So yes, you'll take a knock but that is open to what game you play and what your performance level is prior to running physx. If it was good and the knock not too shabby resulting in the game still being playable then it's trivial imo.

The difference seems to be a little bit more profound than I had originally thought. It also bring up the point that the Physx acceleration is useless on any decent CPU, a good quadcore or an overclocked dualcore. It seems to me if Physx acceleration results in even a slight hit then it is useless for its intended purpose, but I suppose it would be better at lower resolution where the CPU is more important and/or if the CPU is somewhat low-end, is that the consensus here?

I was under the impression that because the majority of systems are limited by the CPU this was meant to help alleviate that and shift some of the processing to the GPU. If this is the case then of course physx makes perfect sense. We see with many benchmarks that graphics cards hit a brick wall where they shouldn't, and this can often be attributed to a CPU that's not able to keep up.

I might be totally off, but that was my understanding.

I assumed this was the reasons why the GTXxxx series dies were soo large. They incorporated Ageias PPU chip. On the other hand the 8xxx series was made before Nvidia bought Ageia.

He wont see much happen.
Im running a 3.2G AMD and a single 8800GTX with Physx on and play crysis all the time.
I have no problems and would guess on my setup it may amount to about 5 fps.

Whats you fps in Crysis? Or, what % is that 5fps loss?

If I remember correctly its around 25 with physx and with the shadows turned down a little.
I will say not everything is on the highest possible as some things I cant see the difference so....
I cant check right now but I will post back soon.

The problem with this is that it assumes current CPUs are powerful enough to run PhysX content on the CPU. They aren't. My i7 gets at best 15-20fps on PhysX enabled scenes (I have 4870 quadfire, so no GPU PhysX acceleration), and even that speed is rare (10fps is more common). It does only use one thread though, so CPU PhysX would be a definite possibility if they would bother to multithread the software implementation.


Oh, and pat - there shouldn't be any difference in Crysis performance between PhysX on and off. Crysis does not use PhysX.

cjl, Im thinking W7 brings this closer to reality, MT in physics using a cpu, tho I may be wrong

That is what I mean, not even my old E8600 ever hit 100% use during gaming alone and my i7 barely budges so there is plenty of power left to spare. I hope that Win 7 will help alleviate this.

I think he said the opposite... that PC's aren't actually powerful enough. I know in certain games my CPU is running at 100% on two of the cores.

Just keep in mind that the CPU can run physics apps, but you can't compare it to a PPU type application or GPU accelerated. It won't boost results necessarily and the hit you see is quite misleading. The reason for this is that using physx; more is done by the physx side in terms of calc per object etc. You get more physics effects and a small* hit in performance. If you had to swing that around onto a CPU then the CPU will surely suffer. The physics the CPU does in a game vs physx is two different matters. That's why Ageia didn't get such a huge market reception - you are trying to sell a product that "seems" to lower performance when applying bigger physic calcs and it takes a good eye to see the effect. Honestly, how can you distinguish between PPU driven physics and CPU driven if it isn't pointed out to you in a game scene?

*Small is open to debate and purely context based. If I do 70FPS before enabling physx and drop 20 by doing so but I can sense the difference in the game "feel" then I'm satisfied with the result.

physx only works with games that support physx which crysis does not crysis has physics but not supported by physx on game that do support physx there will be a small decrease in performance depending on type of video card you have