NVIDIA Makes PhysX Open-Source

[phoronix]

Phoronix: NVIDIA Makes PhysX Open-Source

As a very big surprise bundled alongside the announcement today of the $2,499 USD TITAN RTX graphics card is word that NVIDIA's PhysX software is going open-source!..

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-Open-Source-PhysX

[TemplarGR]

This is nice. Hopefully now we can:

a) Make the cpu-based physx implementation support more threads and better instruction sets, instead of being locked into 1 core and only x87

b) port it to OpenCL, so it can run on non-Nvidia hardware...

[Brisse]

PhysX was sort of "open sourced" in 2015, so what changed exactly? Different licence?

[birdie]

Great!

But it looks like it depends on CUDA which is proprietary and AFAIK only works on NVIDIA GPUs.

[boxie]

PhysX was sort of "open sourced" in 2015, so what changed exactly? Different licence?

was it behind a EULA before? it seems that you needed a developer account before to see the repo (at least according to their site) so it might be a little more open

[Brisse]

was it behind a EULA before? it seems that you needed a developer account before to see the repo (at least according to their site) so it might be a little more open

That may have been the case. I also seem to recall that if you wanted the latest version with all it's bells and whistles then you still had to opt for a closed source PhysX.

[GruenSein]

This is nice. Hopefully now we can:

a) Make the cpu-based physx implementation support more threads and better instruction sets, instead of being locked into 1 core and only x87

b) port it to OpenCL, so it can run on non-Nvidia hardware...

While that would be helpful, it seems almost like starting from scratch. Why not just contribute to Bullet Physics?

[L_A_G]

It would be nice if someone took this and went ahead and created a vendor-neutral GPU implementation using OpenCL or Vulkan.

However knowing Nvidia I doubt they'll be accepting any outside code contributions and particularly not ones that mess with their vendor lock-in. It'll just be their main branch and a bunch of disorganized branches, most of them closed source, maintained by individual companies for their own internal use.

[tuke81]

This is nice. Hopefully now we can:

a) Make the cpu-based physx implementation support more threads and better instruction sets, instead of being locked into 1 core and only x87

b) port it to OpenCL, so it can run on non-Nvidia hardware...

Point a) has been multithreaded and using sse since physX 3...

[schmidtbag]

We’re doing this because physics simulation — long key to immersive games and entertainment — turns out to be more important than we ever thought. Physics simulation dovetails with AI, robotics and computer vision, self-driving vehicles, and high-performance computing.

More like "we're doing this because we spent all this money on a technology where the only real benefit came from people buying our stuff, and as a result, completely lost everyone's interest. But, we're not interested in putting in the effort to support non-Nvidia hardware".
Although locking down PhysX to only Nvidia hardware definitely didn't help its success, I actually don't think that's what led to its unpopularity. As far as I'm concerned, Nvidia made 2 big mistakes with PhysX:
1. The drivers were bloated, and even if you wanted to do AMD for primary graphics and Nvidia for PhysX, getting the drivers installed properly was a headache.
2. Software had to explicitly support GPU-accelerated PhysX. So, even if you had a PhysX-ready system, there was a lot of PhysX-enabled software that wasn't GPU accelerated. So at that point, why not just use a competitor like Havok?