How much load on dedicated PhysX GPU?

Hey all,

I recently picked up a GTX680, and when I did, I used an old GTX460 768MB I had kicking around as a dedicated Phys X board.

I only have two games (at least that I am aware of) that support Phys X. My reason for running Phys X is that Red Orchestra 2 does a lot of bullet physics, and does not multhithread well at all, so it tends to choke even the fastest CPU's.

So I installed the card, told the unified 301.24 to use the GTX460 as a dedicated PhysX board and fired RO2 up.

I proceeded to monitor GPU activity using EVGA's Precision X, and I never saw as much as a blip of activity on GPU2. Disconcerted, I figured that maybe this was because RO2 can be a little buggy, and maybe it wasn't enabling PhysX.

So I decided to test, and fire up Metro 2033 (my other physX supporting game). I played through the entire opening scene (until one of those big flying things knocks over the armored car) and then quit back out to the desktop and reviewed the activity charts. The most GPU use the GTX460 saw was ~2-3%.

So my questions for those of you more familiar with PhysX are as follows:

1a.) What GPU do you use for dedicated PhysX?
1b.) What GPU useage do you see when running PhysX enabled titles.
2.) Is a GTX460 complete overkill for PhysX, or is my system somehow just not using it properly?
3.) Is there a way to test if proper hardware acceleration of PhysX is indeed working?
4.) Is there a way to definitively tell if it is actually working in game?

Thanks,
Matt

1a) i have a gt430
1b) just metro2033 right now (when i use the benchmark application and view the gt 430 usage in GPU Shark, it never gets past 15% usage on the core)
2) if you read the review for hd6970 + gt 430 hybrid physx, it compares how the 460 performs compared to the 430
3) you can tell if it's working/enabled by opening up gpu-z, i have also used GPU Shark from within FurMark to monitor usage. also, try out FluidMark to measure Physx performance
4) i can tell with the metro2033 benchmark, most definitely, lighting is different, not as much on the screen when stuff blows up, etc

edit, here is the thread with the comparisons of gt 430 to gtx 460 when it comes to physx.

http://hardforum.com/showthread.php?t=1645682

looks like it doesn't really helped much and also the gt 430 can come in half height and also doesn't need any extra power connector, it seems to also use 0w when idle

bigted said:

1a) i have a gt430
1b) just metro2033 right now (when i use the benchmark application and view the gt 430 usage in GPU Shark, it never gets past 15% usage on the core)

Click to expand...

Thank you, this is very useful!

The GT430 has 96 cores at 700mhz vs the 460's 336 cores at 675mhz, which means the GTX460 (since the cores are the same design) is 3.375 times faster.

If ou divide your 15% by 3.375 its 4.44% which would be the predicted load on a GTX460. That is within the margin of error for this crude of an estimate for the numbers I ma seeing. Looks like the GTX460 is a bit overkill after all.

bigted said:

2) if you read the review for hd6970 + gt 430 hybrid physx, it compares how the 460 performs compared to the 430

Click to expand...

I will have to check this out. My guess is not very different at all

bigted said:

3) you can tell if it's working/enabled by opening up gpu-z, i have also used GPU Shark from within FurMark to monitor usage. also, try out FluidMark to measure Physx performance

Click to expand...

Thank you, this is very helpful as well.

bigted said:

4) i can tell with the metro2033 benchmark, most definitely, lighting is different, not as much on the screen when stuff blows up, etc

Click to expand...

Thanks,

I was referring more to a indicator "PhysX On" or something like that rather than looking for visual differences which can be subjective, but if you think the difference is that noticeable, then I will take a look at it and see what I find.

bigted said:

edit, here is the thread with the comparisons of gt 430 to gtx 460 when it comes to physx.

http://hardforum.com/showthread.php?t=1645682

looks like it doesn't really helped much and also the gt 430 can come in half height and also doesn't need any extra power connector, it seems to also use 0w when idle

Click to expand...

Thanks for the info.

I just happened to have the 460 which is why I used it. Looks like if you are shopping for a new card to fill the PhysX role, the GT430 is plenty sufficient, costs less, makes less noise and uses less power.

just for the point of thoroughness, this is the exact card i have:
http://www.newegg.com/Product/Product.aspx?Item=N82E16814133432&Tpk=pny gt 430 2gb

i found it at staples on clearance cheaper than newegg, the heatsink will hang down a little into the next PCI/PCI-e slot, luckily my 3rd pci-e x16 slot was at the bottom of my board and did there was plenty of room at the bottom of my case. i also saw minimal loss when my second 6850 went from x16 to x8

as far as a "Physx ON" button, that would be in the settings of the particular game, and in GPU-z there will be a checkmark next to physx

To follow up, I did some more testing last night.

Since Metro 2033 is a single player game (and thus doesn't need 60fps) and it can look really pretty, I upped my graphics settings to the point where I was getting about 30fps.

I think this is why I was recording such low utilization on the PhysX card. The game was being limited by the video card at 30fps.

So I went back in, lowered settings and reran the benchmark over and over again. The highest reading I was able to get on the PhysX card was when video settings were set to 1024x768, and the framerate was in the 80fps range. When I did this, the GTX460 recorded 10-11% PhysX load.

Then I proceeded to do the Fluidmark test By changing all different settings, trying to challenge the dedicated PhysX GTX460 as much as I could, the highest load I got on the card was 61%.

So it's pretty clear to me that the GTX460 is WAY overkill for a PhysX card. In Fluidmark it is probably good even with dual 680's doing the graphics rendering. In regular gaming, one GTX460 is probably overkill even if you have a quad GTX680 system (if you have enough PCIe slots to pull this off)

um Metro 2033 barely uses any physx effects at all. even the normal physx setting for Batman AC uses many times the physx effects being done in Metro 2033. with a gtx680 as your main gpu, a gtx460 is really not overkill for physx. in fact its the recommended card for physx in Batman AC.