NVIDIA Begins Funding Blender Development

[phoronix]

Phoronix: NVIDIA Begins Funding Blender Development

NVIDIA is the latest high profile company now contributing significant funds for advancing the open-source Blender 3D modeling software...

http://www.phoronix.com/scan.php?pag...unding-Blender

[tildearrow]

Bad news!

I wished AMD could fund improvements to OpenCL instead...

[stormcrow]

Bad news!

I wished AMD could fund improvements to OpenCL instead...

While it would be nice, I think that boat has already sailed. CUDA is the GPGPU API of choice in most industries including graphics rendering development. That said, Blender already does support OpenCL very well.

[V1tol]

I think it is better to look into Vulkan's Compute Shaders and move OpenCL and CUDA into maintenance state. That will be more futureproof but I think nVidia won't allow this.

[ermo]

While it would be nice, I think that boat has already sailed. CUDA is the GPGPU API of choice in most industries including graphics rendering development. That said, Blender already does support OpenCL very well.

One can hope that once Vulkan moves forward with its adoption/inclusion of OpenCL(-ish features), that particular barrier can more easily be scaled? Right now, the story around getting full-fat OpenCL drivers working w/AMD's FLOSS drivers seems ... convoluted?

[[email protected]]

OpenCL might still be the future from what I know, as it allows targetting multiple classes of HW accelerators. I also see a lot of OpenMP around these use-cases.

[stormcrow]

One can hope that once Vulkan moves forward with its adoption/inclusion of OpenCL(-ish features), that particular barrier can more easily be scaled? Right now, the story around getting full-fat OpenCL drivers working w/AMD's FLOSS drivers seems ... convoluted?

Perhaps. But it's difficult to overturn an entrenched incumbent. Code doesn't disappear. Companies aren't going to move existing work to Y when they have considerable investment in X. See COBOL for an example. It's relatively easy to unseat an incumbent when you have a mostly drop-in replacement available: Intel x86-64 v. AMD64. It's not so simple when that's not the case Nvidia CUDA v. AMD APP/OpenCL. They aren't direct replacements of each other.

While new code bases might be implemented on whatever Vulkan derivative APIs can offer in the future, existing code may never be ported so Nvidia will have to support CUDA more or less indefinitely in some way or fashion.

Also, there's developer inertia to consider. Professionals tend to familiar tool sets over and over again eg: Unix-like, Windows NT, C, assembly, etc even if they may not be the best choice for the job.

[microcode]

Bad news!
I wished AMD could fund improvements to OpenCL instead...

They already do, to some extent. It's good whenever a vendor does this sort of thing.

[cb88]

While it would be nice, I think that boat has already sailed. CUDA is the GPGPU API of choice in most industries including graphics rendering development. That said, Blender already does support OpenCL very well.

Ah yes... just like Glide, PhysX (good luck finding hardware accelerated PhysX in any new game, it's all slow CPU accelerated junk) and what have you... Proprietary APIs are always doomed even if they succeed temporarily they are doomed in the long run.

[sarmad]

That's nice of nVidia. However, opening their secrets to Nouveau developers would be nicer and more appreciated.