nVidia (almost) opens CUDA

[Kaotik]

http://pressroom.nvidia.com/easyir/...prid=831864&releasejsp=release_157&xhtml=true

http://www.hpcwire.com/hpcwire/2011-12-13/nvidia_opens_up_cuda_compiler.html?featured=top

http://www.anandtech.com/show/5238/nvidia-releases-cuda-41-cuda-goes-llvm-and-open-source-kind-of

They're making it shared source, kinda like Microsoft has plenty of stuff, so in theory AMD, Intel etc could get CUDA from nVidia to allow their own products to support CUDA too - if nVidia decides they're good enough applicants anyway.

 

[Davros]

cant see it as amd could of supported physx (theoretically) but they didnt want to and I cant see them wanting to support cuda

ps: apart from jc2 (uses cuda for waves and I think rage uses it for transcode) any other games use it

 

[Kaotik]

cant see it as amd could of supported physx (theoretically) but they didnt want to and I cant see them wanting to support cuda

ps: apart from jc2 (uses cuda for waves and I think rage uses it for transcode) any other games use it

IIRC the AMD / PhysX/CUDA case wasn't as simple as "didn't want to"

 

[rpg.314]

Thread title needs to be renamed. It's virtually zero opening up.

 

[Kaotik]

Does anyone know what's the current situation on this?
http://nvidianews.nvidia.com/News/NVIDIA-Contributes-CUDA-Compiler-to-Open-Source-Community-7d0.aspx

This was the latest I've found, is CUDA free for anyone (AMD, Intel etc) to adopt for their GPUs now via LLVM backend or not?

 

[mavere]

LLVM also has an ARM backend, but that doesn't mean you can use it to run ARM-specific code on x86.

Here's the NVPTX LLVM backend. It's still Nvidia-only, but the implementation is permissively licensed, so feel free to make new compilers for Nvidia's GPUs.