Any plans or possibilities open sourcing CUDA SDK? GPL or LGPL Licence all CUDA libraries?

What’s the chance of NVIDIA someday making the CUDA SDK fully open source under GPL/LGPL for non-commercial use? Maybe even making it a project, (like Intel, CyberLink, and even Microsoft done for a few of their SDKs)? NVIDIA® Corporation could still own and project manage a such project on or their own website but other independent and hobby developers should be able volunteer (free workers, hehe) if they can prove their skill by submitting patches and/or bug-fixes. Project like that really draw in a big community, people can help writting manuals (online wiki), you could then also use code from other GPGPU programming projects/resources if they too are licenced under GPL/LGPL.

…is that just wishfull thinking? :rolleyes:

PS! I think GPGPU (General-Purpose Computation on GPU) will become a huge deal within only a year or so!

BUMP! No reply yet from the NVIDIA staff :( …not even a no or not at this time answer :ermm:

How about a yes or not/not yet answer about open sourcing the BLAS or FFT library?

Just to start up a basic open source community base around CUDA :hug:

Sorry, no answer at this time.

PS: I think GPGPU has been a huge deal for a while now. But I’m biased. :)

I’d like an answer to this question. I’m considering various high performance computing platforms: AMD/ATI CTM, NVIDIA CUDA and IBM Cell, both CTM and Cell have an open source development environment. Why is NVIDIA keeping the source closed?

We are going to release the sources for CUBLAS and CUFFT soon and you can download the changes to the Open64 compiler today.

Plus, there is a full description of the PTX layer in 1.0, so if you want you can write your own compiler/toolchain.

BTW, where is the source code for the Cell version of xlc :-) ?

I’m not familiar with CUDA architecture, could you please tell me what is being kept closed?

Is there any possibility of opening the card<->OS communication for launching programs?

In a server environment, loading the entire Nvidia driverset is overkill for only running CUDA programs.

Good news about cublas and cufft!

CTM’s SourceForge project continues to show 0 files for download, so I’m not sure that’s a fair comparison.

Asking for the drivers or the driver <-> card interface to be open-sourced is, I suspect, futile. There’s a long history of gfx companies not revealing details about their drivers, probably to protect trade secrets as well as IP that they have licensed that they don’t have the right to open up themselves even if they wanted to.

Some more realistic targets for openness/open source:

  1. Not just a Open64 compiler but buildable sources - NVIDIA seems to be following Pathscale’s practice of releasing the source tree but not the upper level Makefiles or any build instructions. I suppose this is more than coincidence, as NV probably has the same Makefiles. The ideal outcome would be a merge with the Open64 project but in the interim, a build procedure that can work with the Open64 Makefiles would be much appreciated.

It’s understood that a flat source code release is in keeping with the letter of the GPL, but releasing a working build procedure is a little closer to the spirit of the GPL, for what it’s worth.

  1. ptxas?

Raised in a different thread - too much of a new front being opened here. :-)


BUMP! AMD/ATI has now put some presure on NVIDIA by going out an publicly announced that AMD/ATI will from now on officialy do all it can to support the open source community in regards to open source device drivers, API, and SDK. AMD/ATI has begun this campaign by given out full specification (without NDA) for ATI R500 GPUs and above to the driver developers, (and as I understand it they will also soon release the specifications for the R600 media decoder).

John Bridgman and Matthew Tippett are announcing AMD’s open source plans, specifications without NDA. 2D very soon, 3D coming when the legal issues have been taken care of. Specs for the r3xx core will probably be coming later on down the track.

it might take 6-months to a year before end-users see the big difference this will make but there is no doubt that soon both the open source and closed source binary drivers for AMD/ATI graphics chips will soon be better than todays NVIDIA respective device drivers.

So, is the CUDA GPU SDK open or not?
If it is open - can you please post the link to the CUDA GPU SDK source?

John W.

Is this the right forum to ask ^^^?