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

[Viruzz]

Hey,
I hope to make a list of cards and answers here, not just for me but for future use, since this will be needed not just for 5XXX series but all future nvidia generations.

so im looking into buy some cheap, preferably no power connector gpu to use for when decide to play older Physx game.
Any opinion here?
Resolution is 4K/120

Will something like GTX1030 2Gb or 1050Ti 2Gb be enough, im sure it WILL be faster than CPU, but playable faster or 2fps be 10fps?

3050 is perfect but expensive for just be thrown in drawer for when its needed, also they all dual slot, its ok, but i can find 1050 single slot on aliexpress

Also i wonder if vRAM is important? there are 2Gb and 4gb versions of 1050Ti



I found this older video, but it tests vs Single Nvidia or nvidia + dedicated Physx
what we need is something like 5080/5090 + CPU Physx vs 5080/5090 + GPU PhysX [preferably on cheap, no power connector, up to dual slot cards]
Based on this video even 750Ti makes 1070 faster, but 1070 is not a trash level card [for the period] eitehr
BTW is 1050Ti faster than 750Ti?

 

[Solaris17]

Hey,
I hope to make a list of cards and answers here, not just for me but for future use, since this will be needed not just for 5XXX series but all future nvidia generations.

so im looking into buy some cheap, preferably no power connector gpu to use for when decide to play older Physx game.
Any opinion here?
Resolution is 4K/120

Will something like GTX1030 2Gb or 1050Ti 2Gb be enough, im sure it WILL be faster than CPU, but playable faster or 2fps be 10fps?

3050 is perfect but expensive for just be thrown in drawer for when its needed, also they all dual slot, its ok, but i can find 1050 single slot on aliexpress

Also i wonder if vRAM is important? there are 2Gb and 4gb versions of 1050Ti



I found this older video, but it tests vs Single Nvidia or nvidia + dedicated Physx
what we need is something like 5080/5090 + CPU Physx vs 5080/5090 + GPU PhysX [preferably on cheap, no power connector, up to dual slot cards]
Based on this video even 750Ti makes 1070 faster, but 1070 is not a trash level card [for the period] eitehr
BTW is 1050Ti faster than 750Ti?

your not from the ageia physx times are you? Those cards are vastly over powered. Which ones? All of them.

 

[freeagent]

6600GT would be fine

Just kidding.. I am too much of a snob to plug one into my system.

 

[R-T-B]

I think he means to use alongside his existing 5xxx series though, which AFAIK is not possible.

 

[kiddagoat]

You can add another NV card. In the control panel you select what to use for PhysX.

And you can use really anything for it, a 1030 GT would work. Doesn't need a lot.

I had a BFG Ageia with 128MB of memory.

 

[Ruru]

Something like 1060 maybe? I have a GT 640 as a PhysX card in my 3rd PC.

 

[R-T-B]

You can add another NV card. In the control panel you select what to use for PhysX.

And you can use really anything for it, a 1030 GT would work. Doesn't need a lot.

I had a BFG Ageia with 128MB of memory.

The issue is the driver. He can't have two versions installed at the same time, and 5xxx needs a newer driver that does not support 32-bit PhysX, if thats his goal here.

 

[Dr. Dro]

The issue is the driver. He can't have two versions installed at the same time, and 5xxx needs a newer driver that does not support 32-bit PhysX, if thats his goal here.

I was just about to type this as well. It's of particular interest that pre-RTX GPUs are all on the chopping block and driver support will be discontinued any time now (Maxwell through Volta, so remaining 700, 900, 10 series and Titan V). The future of the GTX 16 series is uncertain as well, despite Turing shaders, it's feature incomplete and markedly inferior hardware to the RTX 20 series counterparts. Once those cards are dropped this means you can no longer use those GPUs as 32-bit CUDA acceleration devices if you wish to continue receiving driver updates.

There's really 3 options:

1. RTX 3050 6 GB
2. Used RTX A1000
3. Used RTX A400

The RTX A2000 technically also fits the same criteria, but it's more powerful and also, more expensive. Not to mention it's a complete waste of this card to just leave it there to run PhysX.

 

[freeagent]

The issue is the driver.

Forgot about that..

 

[Shihab]

The issue is the driver. He can't have two versions installed at the same time, and 5xxx needs a newer driver that does not support 32-bit PhysX, if thats his goal here.

It disallows 32bit CUDA on the 5000 card, but does do so even on pre-5000 card with 5000 installed?
NV's announcement leads me to think that the binaries are there, they are simply not loaded/used on a 5000 unit.

The issues I see here are:
1- They'd need a relatively new card still supported by new drivers. 1000 cards are dangerously close from being demoted to legacy status.
2- Nvidia will most likely be removing the 32bit binaries in future drivers.

 

[R-T-B]

It disallows 32bit CUDA on the 5000 card, but does do so even on pre-5000 card with 5000 installed?

yes because nvidia drivers refuse to work with a different version installed on another card, they get all kinds of weird if you try that. Usually just bluescreen.

And 32-bit CUDA support is removed from the latest drivers. Even 4000 series users experience the same if they update.

 

[3x0]

Even 4000 series users experience the same if they update.

Are you sure? I've only read that the latest drivers don't support 32bit CUDA on RTX 5000

 

[Macro Device]

As of early and mid 2010s, sure, CPUs sucked at dealing with PhysX but the today's million core monsters? I thought you might as well consider the CPU penalty negligible on 12900K onwards. Am I wrong?

 

[3x0]

As of early and mid 2010s, sure, CPUs sucked at dealing with PhysX but the today's million core monsters? I thought you might as well consider the CPU penalty negligible on 12900K onwards. Am I wrong?

9800X3D crumbles when running PhysX. Software optimization is important

 

[R-T-B]

Are you sure? I've only read that the latest drivers don't support 32bit CUDA on RTX 5000

"if they update"

The point is you can't install two versions at the same time, so you have to update if you have any 5xxx series card.

I'm not dead certain on 4xxx under the same drivers, but nearly. My point is do a lot of homework before spending.

As of early and mid 2010s, sure, CPUs sucked at dealing with PhysX but the today's million core monsters? I thought you might as well consider the CPU penalty negligible on 12900K onwards. Am I wrong?

Unfortunately physx still likes a ton more than 8 cores. Even 16 falls flat compared to what a gpu has.

 

[Shihab]

And 32-bit CUDA support is removed from the latest drivers. Even 4000 series users experience the same if they update.

Random results on the web seem to indicate binaries are still there. Although efficacy seems to be hit and miss.

https://www.reddit.com/r/nvidia/comments/1iv2a4c
https://www.reddit.com/r/nvidia/comments/1irxi0t/_/mdof3h4

 

[azrael]

I was just about to type this as well. It's of particular interest that pre-RTX GPUs are all on the chopping block and driver support will be discontinued any time now (Maxwell through Volta, so remaining 700, 900, 10 series and Titan V). The future of the GTX 16 series is uncertain as well, despite Turing shaders, it's feature incomplete and markedly inferior hardware to the RTX 20 series counterparts. Once those cards are dropped this means you can no longer use those GPUs as 32-bit CUDA acceleration devices if you wish to continue receiving driver updates.

There's really 3 options:

1. RTX 3050 6 GB
2. Used RTX A1000
3. Used RTX A400

The RTX A2000 technically also fits the same criteria, but it's more powerful and also, more expensive. Not to mention it's a complete waste of this card to just leave it there to run PhysX.

By your reasoning shouldn't a 3050 work as well?

 

[R-T-B]

Random results on the web seem to indicate binaries are still there. Although efficacy seems to be hit and miss.

https://www.reddit.com/r/nvidia/comments/1iv2a4c
https://www.reddit.com/r/nvidia/comments/1irxi0t/_/mdof3h4

Yeah not sure on the binaries, just some random efficacy/"does it work" reports going around.

 

[spectatorx]

The issue is the driver. He can't have two versions installed at the same time, and 5xxx needs a newer driver that does not support 32-bit PhysX, if thats his goal here.

I'm on laptop with 4070, latest driver installed (572.60) and all physx options are still present in nvidia control panel, also just checked borderlands 2, physx works on gpu as it have always been working there.

The same driver works for both 5xxx and 1xxx gpus so there should be no problem with making, for example, 1050/ti being delegated to physx only aside 5090 running as the main gpu.

If anyone still have any doubts about that then i bet it would be a great idea for test and article for techpowerup team.

 

[Vya Domus]

As of early and mid 2010s, sure, CPUs sucked at dealing with PhysX but the today's million core monsters? I thought you might as well consider the CPU penalty negligible on 12900K onwards. Am I wrong?

PhysX was purposely unoptimized garbage on CPUs and was not multithreaded, later versions ran better but still not that great.

 

[Red Hood]

Nvidia found a solution to all those unsold 3050s

1050 should be ok for a physx card

 

[Klemc]

My 4070ti on 572.60 loads PhysX high in Alice 2, it's just 5000 that are emputed !

 

[Viruzz]

your not from the ageia physx times are you? Those cards are vastly over powered. Which ones? All of them.

Which are overpowered? 5XXX lacks 32bit Physx support and it runs on CPU
So the only way to play older games with PhysX, you need add-on card
Now, if you game in 4K/120 than the load on PhysX is also increased right?

So which card will be ok for that?

It disallows 32bit CUDA on the 5000 card, but does do so even on pre-5000 card with 5000 installed?
NV's announcement leads me to think that the binaries are there, they are simply not loaded/used on a 5000 unit.

The issues I see here are:
1- They'd need a relatively new card still supported by new drivers. 1000 cards are dangerously close from being demoted to legacy status.
2- Nvidia will most likely be removing the 32bit binaries in future drivers.

They cant just remove 32bit binaries since its supported on 4XXXX series and all other RTX cards besides 5XXXX

 

[Shihab]

They cant just remove 32bit binaries since its supported on 4XXXX series and all other RTX cards besides 5XXXX

Supported with current drivers =\= support with future drivers. Nvidia began deprecating 32 bit CUDA several major versions ago. It's probably safe to say that all modern CUDA-based software has switched to 64bits by now. It will eventually be removed entirely, in which case 4000 users who require it (and those would be a very small minority) can simply stick to older drivers.

I suppose Nvidia can be "nice" enough to keep the bits in until Ada gpus receive the legacy status. But it's probably unwise to bet on megacorps being "nice".

 

[PLAfiller]

Someone posted the Linus video I was thinking about, but basically they conlcuded that pairing a new card with a very old card supporting Physx aint' that great. I mean it's best if the "Physx" card is a closer generation to the main card that is used. (at least that's what I remembered). If you have a 5xxx card, let's say something like RTX A2000? They are fairly cheap after mining, no power connector, extremely efficient. 4050/3050 something of the sort IMO.