Recommended PhysX card for 5xxx series? [Is vRAM relevant?]

[b1k3rdude]

@b1k3rdude Here's GPU-z for both cards :

It seems I misread the wikipedia info for the 8800 - https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units

 

[AleXXX666]

9800X3D crumbles when running PhysX. Software optimization is important

don't compare apples to tomatoes. Cache vs cores, lol.

 

[Shihab]

Rejoice, Blackwell beta testers! Soon you'll have a new thing to beta test.


ZLUDA Continues Working On PyTorch Support, Eyes 32-bit PhysX Support

 

[tugrul_LordOfDrinks]

Recommend imagination. Brain is the fastest gpu.

Maybe one day all producers unite and create a standard API for physics in games.

 

[Sydneyblue]

Back in the Day ran a Physics card using GTS 250 and gtx 260 as main card. When i got my 470 GTX there was no reason to use a physics card because the 470 can do a better job being a main card and doing physics. I dont think its Viable. Is the gains worth having another card dumping more heat ? vs gains ? I miss having SLI and multi card set ups but not that much

 

[THU31]

I managed to find a GT 1030 GDDR5 for about 25 bucks, so I got it for testing purposes. It fits nicely in the bottom PCI-E x4 slot (connected to the chipset).

I only tested Batman: Arkham City so far (4K, max settings, MSAAx4).
When I put PhysX just on the 4070, the GPU is underutilized in all PhysX-heavy scenes in the benchmark, even as low as 60%.
When I put PhysX on the 1030, the 4070 is fully utilized in most PhysX-heavy scenes, only the scene with shattered glass shows lower utilization.
The utilization of the 1030 never goes above 50%. Not sure how relevant that is. I assume that PhysX calculations don't use many parts of the GPU (like TMUs, ROPs etc.).

There definitely seems to be some kind of bottleneck, since the 4070 is underutilized when it's the only active card (and the glass scene is always the lowest performing one, including on the CPU).
I'd be interested to see the results with a 5070 Ti, I wonder if there would be a similar bottleneck of some kind. Maybe I'll get to find out later this year.

 

[jarablue]

I managed to find a GT 1030 GDDR5 for about 25 bucks, so I got it for testing purposes. It fits nicely in the bottom PCI-E x4 slot (connected to the chipset).

I only tested Batman: Arkham City so far (4K, max settings, MSAAx4).
When I put PhysX just on the 4070, the GPU is underutilized in all PhysX-heavy scenes in the benchmark, even as low as 60%.
When I put PhysX on the 1030, the 4070 is fully utilized in most PhysX-heavy scenes, only the scene with shattered glass shows lower utilization.
The utilization of the 1030 never goes above 50%. Not sure how relevant that is. I assume that PhysX calculations don't use many parts of the GPU (like TMUs, ROPs etc.).

There definitely seems to be some kind of bottleneck, since the 4070 is underutilized when it's the only active card (and the glass scene is always the lowest performing one, including on the CPU).
I'd be interested to see the results with a 5070 Ti, I wonder if there would be a similar bottleneck of some kind. Maybe I'll get to find out later this year.

That sounds like a ranky dinkle...I hope you get it sorted out!

 

[Ninja Weedle]

I managed to find a GT 1030 GDDR5 for about 25 bucks, so I got it for testing purposes. It fits nicely in the bottom PCI-E x4 slot (connected to the chipset).

I only tested Batman: Arkham City so far (4K, max settings, MSAAx4).
When I put PhysX just on the 4070, the GPU is underutilized in all PhysX-heavy scenes in the benchmark, even as low as 60%.
When I put PhysX on the 1030, the 4070 is fully utilized in most PhysX-heavy scenes, only the scene with shattered glass shows lower utilization.
The utilization of the 1030 never goes above 50%. Not sure how relevant that is. I assume that PhysX calculations don't use many parts of the GPU (like TMUs, ROPs etc.).

There definitely seems to be some kind of bottleneck, since the 4070 is underutilized when it's the only active card (and the glass scene is always the lowest performing one, including on the CPU).
I'd be interested to see the results with a 5070 Ti, I wonder if there would be a similar bottleneck of some kind. Maybe I'll get to find out later this year.

I actually have a 5070 Ti and a 3050 6GB hooked up in my machine now, I haven’t taken a look at any physx games yet though.

 

[THU31]

I actually have a 5070 Ti and a 3050 6GB hooked up in my machine now, I haven’t taken a look at any physx games yet though.

Do you have Batman: Arkham City? It would be nice to have a direct comparison with the built-in benchmark.

It would also be relevant to know what CPU you have. The bottleneck I encountered is definitely CPU-related. At 5.1 GHz the minimum framerate was 70 FPS, but at 3.3 GHz it was 59 FPS.
Only the scene with shattered glass seems to be affected. It seems that some calculations are being done on the CPU, even if you're accelerating PhysX with either the main GPU or a dedicated one.

I think I remember Mafia II having a similar issue, where certain elements (cloth physics?) were calculated on the CPU.

EDIT.

My friend has a 7800X3D and a 4090. He got average 143, max 223 and minimum 78, so there's definitely a CPU bottleneck even when hardware accelerated.

 

[Ninja Weedle]

Do you have Batman: Arkham City? It would be nice to have a direct comparison with the built-in benchmark.

It would also be relevant to know what CPU you have. The bottleneck I encountered is definitely CPU-related. At 5.1 GHz the minimum framerate was 70 FPS, but at 3.3 GHz it was 59 FPS.
Only the scene with shattered glass seems to be affected. It seems that some calculations are being done on the CPU, even if you're accelerating PhysX with either the main GPU or a dedicated one.

I think I remember Mafia II having a similar issue, where certain elements (cloth physics?) were calculated on the CPU.

I do have Arkham City (Haven't played it yet). Using a Ryzen 9700X at stock clocks, 4K Max settings with 8x MSAA with the game running on a 7200RPM HDD:
3050 enabled:
11 minimum
197 Maximum
154 Average

Performance is consistent throughout, and doesn't drop at all even in the glass scene.

All the stutters here I believe were storage related as they occurred when switching scenes, PhysX has no issues thanks to the 3050.

3050 Disabled:
2 Minimum
186 maximum
103 Average

As you might predict, the glass scene is bad. It dips as low as 48 FPS there and averages about 50, and there's noticeable breaks in the water streams compared to Hardware PhysX. Scenes with smoke were also affected, with dips to 78 in the sparks scene and 64 in the room where the grunt runs through the smoke.

So yeah, if you want to play the Arkham trilogy on 50 series, You'll want a PhysX card. Fun!

 

[THU31]

Thanks for testing. The game has shader compilation stutter, by the way, both the benchmark and the actual game stutter a lot without any cached shaders.
But according to pcgamingwiki, running it from an HDD also causes stuttering supposedly.

Interesting that you have high performance in the glass scene with a 3050. Maybe this particular scene does require a lot more computational power for PhysX.