NVIDIA have now made PhysX open source

If anyone was confused, like me, about this news, the earlier 'opening' (-not-really-though) of PhysX was this: https://developer.nvidia.com/physx-source-github , when they made it available over github to 'registered developers'. (I don't know what the latter implies exactly.)

It looks like this time they're opening it up for good, under a BSD license.


Last edited by walther von stolzing on 3 December 2018 at 2:27 pm UTC

So now engine projects based on GPL code from say id Software should be able to use PhysX? Then that is great news.

Edit:

So the upcoming 4.0 release will also be BSD licensed?


Last edited by Kristian on 3 December 2018 at 2:48 pm UTC

If anyone was confused, like me, about this news, the earlier 'opening' (-not-really-though) of PhysX was this: https://developer.nvidia.com/physx-source-github , when they made it available over github to 'registered developers'. (I don't know what the latter implies exactly.)

FWIW anyone could register there and access it. Even I have an account there and I'm no developer.

If anyone was confused, like me, about this news, the earlier 'opening' (-not-really-though) of PhysX was this: https://developer.nvidia.com/physx-source-github , when they made it available over github to 'registered developers'. (I don't know what the latter implies exactly.)

FWIW anyone could register there and access it. Even I have an account there and I'm no developer.

I guess the difference is that this is the new (4.0) SDK, and it's on the BSD aka 'do whatever you want with it' license.

I also didn't know that physx has applications in robotics etc.; that makes it more interesting, because frankly, the effect in games isn't *that* big of a deal imho. (At least it wasn't in titles like Arkham City or Assassin's Creed 4)

does this mean AMD can implement GPU physx?

NVIDIA and open source in one sentence? What?

Now someone "only" needs to migrate that code over to use OpenCL instead of CUDA, leading to a fully portable physics system.

....and? How many linux games supports Physx by GPU?

Better late than never.

Well, I guess theyll make Raytracking opensource when its not relevant(a gimmick to sell videocards) anymore.

It doesn't look too useful. Still tied to CUDA for hardware acceleration.

Also, what stops anyone from using Vulkan compute shaders instead of OpenCL?


Last edited by Shmerl on 3 December 2018 at 5:30 pm UTC

Maybe now Borderlands can be patched with PhysX support? I know that was the main difference between the Windows and Linux versions.

And I had trouble with Mesa's OpenCL implementation, which is a shame.

Mesa's Clover OpenCL state tracker is a bit outdated. The main reason is that AMD is now focusing on their ROCm platform, which sadly is not yet 100% open source. It currently needs some closed source components for OpenCL image support, but AMD said that those components will either be open sourced or replaced in the near future.
Nevertheless, I just recently installed it via the unofficial Gentoo ebuilds, and I must say, that it indeed ROCks.

does this mean AMD can implement GPU physx?

Yes. Someone could also extend OpenCL, etc. versions of it and make it even MORE generic.

Well, I guess theyll make Raytracking opensource when its not relevant(a gimmick to sell videocards) anymore.

Meh... One could get there if you followed the papers. Patents MIGHT be of concern, but they'd need to have filed those patents within a year of their disclosure of the basics of the whole process.

does this mean AMD can implement GPU physx?

Yes. Someone could also extend OpenCL, etc. versions of it and make it even MORE generic.

doesnt nvidia has a processor thingy in the gpu, which is only for physx? so it isnt just software, but hardware?

Nevertheless, I just recently installed it via the unofficial Gentoo ebuilds, and I must say, that it indeed ROCks.

Just got me a RX 580 and was sad to learn that I can't use ROCm because ... they think my i7 2600k Sandy Bridge is too old!
So OpenCL on my GPU does not work because of my CPU
That leaves me with Clover and OpenCL 1.1, worse than with my old Nvidia Card before, not exactly what I was expecting...

I think Bullet was going for that, but last I heard it was still experimental. And I had trouble with Mesa's OpenCL implementation, which is a shame.

Yeah, it's a bit of a pain. Maybe someone can get it sorted out, or maybe get an impetus to get someone ON that. This being there as a largely full and complete framework would be a start. Sometimes it's better to work with something more full than "Open Source" to get a goad applied to get something done.

Of course, compute is available through OpenGL and Vulkan, but I don't think they're as complete (for lack of a better term) for a purely compute based solution like OpenCL or CUDA. If the physics was just for gaming, not sure that would matter.

"Complete," isn't a word I'd use for that. OpenCL makes it easier to code for a given GPGPU compute kernel, but saying it's incomplete is not strictly correct. You're doing odd, counterintuitive operations within textures, surfaces, etc. to get compute operations in OpenGL or Vulkan.

But, I'm sure someone will start converting from cuda (if this release includes the gpu acceleration - I've not delved into it, but maybe it's just the CPU side of physx).

Heh. Might want to go over to the linked Blog page.

"Free, Open-Source, GPU-Accelerated"

I think they're giving it ALL out.


Last edited by svartalf on 3 December 2018 at 7:09 pm UTC

Nevertheless, I just recently installed it via the unofficial Gentoo ebuilds, and I must say, that it indeed ROCks.

Just got me a RX 580 and was sad to learn that I can't use ROCm because ... they think my i7 2600k Sandy Bridge is too old!
So OpenCL on my GPU does not work because of my CPU
That leaves me with Clover and OpenCL 1.1, worse than with my old Nvidia Card before, not exactly what I was expecting...

Uhm... WHAT? The only real differences there is overall speed of the CPU and the Intel GPU (which isn't getting used...or, rather, I hope not..). There's no architectural or ISA differences that should matter or prevent it from being usable on that CPU.

does this mean AMD can implement GPU physx?

Yes. Someone could also extend OpenCL, etc. versions of it and make it even MORE generic.

doesnt nvidia has a processor thingy in the gpu, which is only for physx? so it isnt just software, but hardware?

Nope. They bought the company which had dedicated silicon before they got bought- which is part of WHY they got bought. Another $150-200 card for physics accel? It didn't sell well. Once NVidia bought them, they supported the old, now utterly deprecated, hardware for a while, but moved it all over to CUDA compute kernels. All but the lowest end cards make the other hardware utterly irrelevant- because it had more moxie than the cards ever imagined of having.