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.

 

[ShrimpBrime]

I don't like the idea of testing PhysX speed in old Vantage... that version was a mess IIRC.

From practical point of view though, regardless of how much faster G80 actually is, I'd say GT 710 is the better PhysX accelerator of the two.
1) Smaller (single slot/half height usually)
2) No additional power plugs (and less heat dumped into your case)
3) Newer driver support (not sure if newest standalone PhysX can run on G80, OR if is/isn't nerfed vs. old versions due to lack of optimization)
4) It shouldn't die as easily due to... suboptimal manufacturing techniques of the time.

@b1k3rdude Here's GPU-z for both cards :
View attachment 391734View attachment 391735

Well it turns out the Ageia physx does work with current version of Vantage.

No NVidia GPU installed, this is Rare PPU2 Ageia Physx card that was never released to the public. I did work up the nerve to dig it out and fire it up. DDU all NV drivers, uninstall and manually remove all NVidia Physx drivers. Using the last actual Ageia driver version 7.11.13 comes with 3 demos. Most of the other demos I've tried do not work well with W11 or other file issues that prevent the correct operation of said demos.

However, there is a benchmark that seems to run with or without an Ageia Physx card maybe some of you guys would be interested in.
can down load the "Ageia RealityMark PPU Benchmark" From Vogons here: https://vogonsdrivers.com/getfile.php?fileid=967&menustate=0

I actually have the game installed, but some file is incorrectly installed so I can only play the maps that support no Ageia physx adapter. I can't quite remember how to get it working properly, but it runs pretty good in software mode. It's a lot, I mean a whole lot smoother than it was back in 06. Processors have come quite a long way. https://vogonsdrivers.com/getfile.php?fileid=949&menustate=0

Here's the screeny of 3DMark Vantage and the PPU2,14900K and 7900 XT



A couple Demo Videos.
Edit
additional comment:
It still looks better than software physx. Tearing the cloth is very realistic simulation even with a mouse cursor. 2 decades old, still looks cool as F imo. (1080P)

 

[Ninja Weedle]

I DON'T RECOMMEND GT 1030s/P620s FOR PHYSX ANYMORE WITH RTX 50 SERIES, THEY'RE TOO SLOW (And also don't use PCIE x1 for it)

I was talking to a guy on Reddit recently who wanted to get a GT 1030 for use with his 5090, but he wanted to use it in an x1 slot since that's all he had available. Curious about the bandwidth limitations, I got to testing- after all, these GPUs aren't even hitting 100% utilization with PhysX usually, so what could go wrong? Well, I'm just gonna post my comment verbatim here:

Tested arkham knight on my 5070 Ti and Quadro P620 at 4K MAX with all the PhysX options enabled, here are my results at 3.0 x4 and x1 respectively. Note that the game was loaded on an HDD, and I’m using the built in benchmark:

P620 x4: 40 fps minimum, 107 average

P620 x1: 20 fps minimum, 64 average.

CPU only PhysX, stock 9700x: 56 minimum, 127 average.

3050 PCIE 4 x4: 75 fps minumum, 155 average.

No difference was observed between selecting the 5070 Ti or the CPU as the PhysX processor. I’ve also thrown in results from my 3050 6GB to show what a better secondary GPU does

So in Arkham Knight, it’s actually BETTER to skip the 1030 altogether, and x1 leads to terrible performance. This genuinely surprised me- i remember the framerate tanking in arkham asylum with physx, but knight seems to take it in stride. This is a late 32 but physx title though- what about the earlier arkham games?

Arkham City’s physx usage in the benchmark is particularly brutal, and it tells a different story. Here it is at 4K MAX with 2xMSAA, with the benchmarks run twice to remove loading stutters:

P620 x4: 72 minumum, 139 average

CPU only: 42 minimum, 136 average

3050 PCIE 4 x4: 121 minimum, 229 average

It’s worth noting that this game somewhat scales back the amount of glass particles in one scene when it runs on a CPU compared to a GPU.

Didn’t even bother running this one in x1 since it would easily be worse than CPU. The main difference here is minimums- the biggest downside of CPU PhysX is stutter, and that’s quite apparent here. The 3050 also shows MASSIVE gains here over CPU.

Basically? Bandwidth MATTERS for PhysX, and it matters a lot when paired with a 5070 Ti, and that only increases with something as fast as a 5090. This is also making me rethink my recommendation of a P620 for PhysX at the high end- the gain over CPU is there, but with a modern processor, it's not that much better. It helps with the 1% lows certainly, but a better card helps out with the 1% lows that much more. Adding a second GPU seems to only make sense with considerably faster cards, like the 3050. I need a bigger sample size to determine what exactly is the most important aspect (Is it PCIE bandwidth? Raw card horsepower?) for PhysX though. Makes me wonder what a 5090 + 4090 combo would look like, assuming they don’t catch fire.


Overall? I think it might be better to reccommend GTX 1650/Supers, 1060s, and RTX 3050s for PhysX rather than the old low profile pascal cards. Cards like the GT 1030 just aren't fast enough to not hold back 50 series cards in these games substantially, and may just not be worth the purchase if you already have a great CPU. CPU still sucks, but if you're going to buy a PhysX card, get something > GTX 1050.

 

[ShrimpBrime]

what exactly is the most important aspect

Game optimization is the most important aspect. There is little of that going on.

Best Game to test with that actually ran halfway decent is Hawken. It utilized Physx Destruction particle and cloth physx. It had it's choppiness, but that was years ago and 5070ti's didn't exist yet.

 

[Dr. Dro]

I DON'T RECOMMEND GT 1030s/P620s FOR PHYSX ANYMORE WITH RTX 50 SERIES, THEY'RE TOO SLOW (And also don't use PCIE x1 for it)

I was talking to a guy on Reddit recently who wanted to get a GT 1030 for use with his 5090, but he wanted to use it in an x1 slot since that's all he had available. Curious about the bandwidth limitations, I got to testing- after all, these GPUs aren't even hitting 100% utilization with PhysX usually, so what could go wrong? Well, I'm just gonna post my comment verbatim here:

Tested arkham knight on my 5070 Ti and Quadro P620 at 4K MAX with all the PhysX options enabled, here are my results at 3.0 x4 and x1 respectively. Note that the game was loaded on an HDD, and I’m using the built in benchmark:

P620 x4: 40 fps minimum, 107 average

P620 x1: 20 fps minimum, 64 average.

CPU only PhysX, stock 9700x: 56 minimum, 127 average.

3050 PCIE 4 x4: 75 fps minumum, 155 average.

No difference was observed between selecting the 5070 Ti or the CPU as the PhysX processor. I’ve also thrown in results from my 3050 6GB to show what a better secondary GPU does

So in Arkham Knight, it’s actually BETTER to skip the 1030 altogether, and x1 leads to terrible performance. This genuinely surprised me- i remember the framerate tanking in arkham asylum with physx, but knight seems to take it in stride. This is a late 32 but physx title though- what about the earlier arkham games?

Arkham City’s physx usage in the benchmark is particularly brutal, and it tells a different story. Here it is at 4K MAX with 2xMSAA, with the benchmarks run twice to remove loading stutters:

P620 x4: 72 minumum, 139 average

CPU only: 42 minimum, 136 average

3050 PCIE 4 x4: 121 minimum, 229 average

It’s worth noting that this game somewhat scales back the amount of glass particles in one scene when it runs on a CPU compared to a GPU.

Didn’t even bother running this one in x1 since it would easily be worse than CPU. The main difference here is minimums- the biggest downside of CPU PhysX is stutter, and that’s quite apparent here. The 3050 also shows MASSIVE gains here over CPU.

Basically? Bandwidth MATTERS for PhysX, and it matters a lot when paired with a 5070 Ti, and that only increases with something as fast as a 5090. This is also making me rethink my recommendation of a P620 for PhysX at the high end- the gain over CPU is there, but with a modern processor, it's not that much better. It helps with the 1% lows certainly, but a better card helps out with the 1% lows that much more. Adding a second GPU seems to only make sense with considerably faster cards, like the 3050. I need a bigger sample size to determine what exactly is the most important aspect (Is it PCIE bandwidth? Raw card horsepower?) for PhysX though. Makes me wonder what a 5090 + 4090 combo would look like, assuming they don’t catch fire.


Overall? I think it might be better to reccommend GTX 1650/Supers, 1060s, and RTX 3050s for PhysX rather than the old low profile pascal cards. Cards like the GT 1030 just aren't fast enough to not hold back 50 series cards in these games substantially, and may just not be worth the purchase if you already have a great CPU. CPU still sucks, but if you're going to buy a PhysX card, get something > GTX 1050.

I tried my RTX A2000, it works... fine, I guess? But I'm honestly not too fussed about it. I reached the conclusion that it was actually better for me to just remove it, let my 5090 have some more air and give up on PhysX. Installed another SSD with the extra lanes available and that was that. I get to keep the 5090 at x16, too.

I didn't have anything useful to do with it so I actually gave my A2000 to my brother so he could build himself a small ITX HTPC with one of his 4700S boards. Dude was using a GTX 980 up until now. He says he'll bring it home for me to see when he finishes it. The king is dead, long live the king. I won't exactly miss it.

 

[THU31]

I DON'T RECOMMEND GT 1030s/P620s FOR PHYSX ANYMORE WITH RTX 50 SERIES, THEY'RE TOO SLOW (And also don't use PCIE x1 for it)

I was talking to a guy on Reddit recently who wanted to get a GT 1030 for use with his 5090, but he wanted to use it in an x1 slot since that's all he had available. Curious about the bandwidth limitations, I got to testing- after all, these GPUs aren't even hitting 100% utilization with PhysX usually, so what could go wrong? Well, I'm just gonna post my comment verbatim here:

Tested arkham knight on my 5070 Ti and Quadro P620 at 4K MAX with all the PhysX options enabled, here are my results at 3.0 x4 and x1 respectively. Note that the game was loaded on an HDD, and I’m using the built in benchmark:

P620 x4: 40 fps minimum, 107 average

P620 x1: 20 fps minimum, 64 average.

CPU only PhysX, stock 9700x: 56 minimum, 127 average.

3050 PCIE 4 x4: 75 fps minumum, 155 average.

No difference was observed between selecting the 5070 Ti or the CPU as the PhysX processor. I’ve also thrown in results from my 3050 6GB to show what a better secondary GPU does

So in Arkham Knight, it’s actually BETTER to skip the 1030 altogether, and x1 leads to terrible performance. This genuinely surprised me- i remember the framerate tanking in arkham asylum with physx, but knight seems to take it in stride. This is a late 32 but physx title though- what about the earlier arkham games?

Arkham City’s physx usage in the benchmark is particularly brutal, and it tells a different story. Here it is at 4K MAX with 2xMSAA, with the benchmarks run twice to remove loading stutters:

P620 x4: 72 minumum, 139 average

CPU only: 42 minimum, 136 average

3050 PCIE 4 x4: 121 minimum, 229 average

It’s worth noting that this game somewhat scales back the amount of glass particles in one scene when it runs on a CPU compared to a GPU.

Didn’t even bother running this one in x1 since it would easily be worse than CPU. The main difference here is minimums- the biggest downside of CPU PhysX is stutter, and that’s quite apparent here. The 3050 also shows MASSIVE gains here over CPU.

Basically? Bandwidth MATTERS for PhysX, and it matters a lot when paired with a 5070 Ti, and that only increases with something as fast as a 5090. This is also making me rethink my recommendation of a P620 for PhysX at the high end- the gain over CPU is there, but with a modern processor, it's not that much better. It helps with the 1% lows certainly, but a better card helps out with the 1% lows that much more. Adding a second GPU seems to only make sense with considerably faster cards, like the 3050. I need a bigger sample size to determine what exactly is the most important aspect (Is it PCIE bandwidth? Raw card horsepower?) for PhysX though. Makes me wonder what a 5090 + 4090 combo would look like, assuming they don’t catch fire.


Overall? I think it might be better to reccommend GTX 1650/Supers, 1060s, and RTX 3050s for PhysX rather than the old low profile pascal cards. Cards like the GT 1030 just aren't fast enough to not hold back 50 series cards in these games substantially, and may just not be worth the purchase if you already have a great CPU. CPU still sucks, but if you're going to buy a PhysX card, get something > GTX 1050.

Isn't it weird that there's no difference between selecting the CPU and the 5070 Ti in Arkham Knight? It's a 64-bit game, so it should be accelerating PhysX even without a dedicated card.

As for Arkham City, it's interesting that a 3050 pretty much doubles your framerate. I guess it makes sense, as I had similar results with my 4070+1030.

1030 should be enough for 60 FPS in older 32-bit games, but for high framerates you'll need something more powerful (which will affect placement of the card and thermals inside the case). Worth noting, that many older games break above 60 FPS, at least partially.

 

[ShrimpBrime]

RTX 5070 - Arkham Asylum - BATMAN!!!