Dedicated 9800GTX PhysX investigated + MAFIA II benchmarks

I decided to investigate how much the processing power of a dedicated PhysX card can affect game framerate - eg Mafia II

It is often said that the number of shader units are key to the PhysX processing capability of the dedicated card. I wanted to check this and also how core and memory speed will affect framerates.

System:

E8500@4.1ghz
GTX470 (OC - 760Mhz)
9800GTX (ded. PhysX)
2GB DDR3 1500Mhz

MAFIA II Demo
PhysX HIGH
1280x1024
16xAF+AA set in game

Testing method:
I used MSI Afterburner to DOWNCLOCK the 9800GTX to the minimum possible speed (505mhz Core, 1263mhz Shader clock, 825mhz Memory) . Then I increased Core, Shader and Memory clocks in turn to see how they affected framerates.
Finally I overclocked the card by 15% to see if I could squeeze some more fps out of it.

Results:





No real surprises. Shader capability is indeed where it's at for PhysX processing. Out of the 3 variables, increasing the Shader clock gives by far the greatest gains. It should be noted that Afterburner would not allow me to leave the Shader clocks at minimum when raising the Core, so the raise in Core clock is less effective than the chart implies.

It was interesting to see that a 15% overclock on the 9800GTX did give significant gains in fps.

Conclusion:
Bump up your Shaders! It is definitely worth overclocking your dedicated PhysX card if you are using a 9800GTX or slower card. I imagine that you will get diminishing returns with GTX275 and upwards but if you are lucky enough to be using such a card for dedicated PhysX, why not? You can leave or even downclock your dedicated GPUs core and RAM in an attempt to get the max shader clock possible.

 

Nice, thanks for posting this, might try the same thing clock wise with my GTX260, and 9600GSO to see the difference.

 

Great work man, thanks for the info!

 

interesting, this thread however, indicates that different cards with different CUDA core counts didn't seem to have any significant effects to the physx performance, it didn't show any info when the physx cards are overclocked though...

 

^^^^wut??

It shows the 9800 oc'd and even broke down the oc'd components?

 

I think its pontless to have gtx470 n older card doing physx ...i did it with 470 n gtx280 as physx n losted 100 in physx fluidmarks (got higher score without 280.n didnt c much of diff in mafia 2 with this setup .if u did on win7 i dont c how ur 9800gtx was able to render dx11 physx since only dx10 the most..
Wont there b a physx rending problem ...even under xp i didnt c much of a physx boost

 

well he seems to be gaining almost an extra 20fps using a oc'd 9800 as physx in mafia2......

 

A dedicated card makes a huge difference. It takes Batman AA/Mafia 2 in 3D from being unplayable to being perfectly smooth on my system. I've even over clocked my PHYSX card because of the information i gained from this thread and noticed similar gains. Respect to OP for his work.

 

a dedicated physx card is a must. im running 480 graphics and a gtx 260 for physx and i get o huge boost over just running my 480. 10-15 fps more in mafia 2 benchmark. can get 42fps avg or higher with all settings on max

 

If you look at my results you will see the big difference that the dedicated PhysX card makes.

The reason you saw a drop in score when using fluidmark is because fluidmark does not max out your GPU. Your 470 had plenty of headroom to do the physx processing.

In gaming, your GPU has better things to do than physics processing and this (as said by the 2 previous posters) is when a dedicated PhysX card is required.

The 9800GTX was doing no rendering, it was doing PhysX processing. The version of DirectX is completely irrelevant.

 

Here is some Mafia II benchmarks, which Card is best for PhysX:





And here a Benchmark what a Geforce GT 430 dos as PhysX Card: