CUDA going Open Source

Hmm, would need for information.... nvidia is releasing the code to "qualified academic researchers and software tools developers"

need to do more reading, but if it is similar to what it sounds like, then nvidia must NOT be liking what AMD/Intel/OpenCL may be working on to make a move like this.
 
Shame on NVIDIA, they could have done this a lot earlier, I don't mean just when they first released CUDA, but 5 years later is too much time.

I have to agree with the previous post saying that AMD may have something big under their umbrella for NVIDIA releasing this source code. As we all know they are reluctant to release code of their closed source drivers for the Linux kernel.
 
Slashdot

"NVIDIA has announced they have 'open-sourced' their new CUDA compiler so that their GPGPU platform can be brought to new architectures. NVIDIA's CUDA compiler is based upon LLVM. At the moment though they seem to be restricting the source code's access to 'qualified' individuals.' The official press release implies wider access to the source will happen later. It so happens that a few days ago AMD opened their OpenCL backend and added initial support to the Free Software r600 driver.
 
This is probably a lateral move for Nvidia as the only advantage Khronos Groups' OpenCL had over CUDA is the fact OpenCL can be used to run on arbitrary hardware (or mix thereof). The fact that Nvidia hardware runs CUDA code about 15% faster then OpenCL always hinted to me that Nvidia was ahead of the curve in compiler tech compared to Khronos guys.

Also, Intel has just now gotten their AVX featured stuff running pretty good (relatively speaking) using OpenCL code (and in such a short time). In that regard, Nvidia is probably more afraid of Intel throwing their weight around towards getting their hardware or future hardware (MIC/Larrabee?) running OpenCL stuff better than Nvidia's own stuff running CUDA. Not happening right now, but never underestimate Intel given their vast resources.

The Portland Group had been working on CPU emulation for CUDA in the past, but it seems they shelved it (or so it seems to outsiders). Having CUDA running on arbitrary hardware is something alot of developers (and not just crazy supercomputing physicists) have been interested in.

And if the rumors are true and AMD's next gen hardware will be ditching the VLIW stuff for the more traditional threading/Warp scheduling stuff that Nvidia has been doing here lately, then there exists a strange possibility that future AMD hardware will be inherently more CUDA optimized, at least theoretically since CUDA doesn't run on any AMD devices right now, period.

So all in all, this is a good thing. Being able to run CUDA code on arbitrary hardware is a win/win for everyone else. Let's just hope Nvidia plays fair by not throwing any curveballs when open sourcing their compiler tech.
 
This is a good move. It's nearly impossible to gain wide acceptance for something that only runs on one hardware vendor.

Now about that Fizz-Ex...
 
2ik5xxf.jpg
.
 
Technikal said:
Slashdot

"NVIDIA has announced they have 'open-sourced' their new CUDA compiler so that their GPGPU platform can be brought to new architectures. NVIDIA's CUDA compiler is based upon LLVM. At the moment though they seem to be restricting the source code's access to 'qualified' individuals.' The official press release implies wider access to the source will happen later. It so happens that a few days ago AMD opened their OpenCL backend and added initial support to the Free Software r600 driver.
Click to expand...
At its heart is CUDA C which is compiled with NVIDIA’s compiler (nvcc – based on the Open64 back-end). To be clear, CUDA C is not C – it is a variant of C with extensions.
 
It's open sourced as in how microsoft does 'Shared Source' that is, you can apply and they may approve you access, but only enough to do what you need to do, not enough to completely figure out how it works or say port it to another card. This is more of a move to assist developers who want to make ARM + Nvidia clusters than a move to cut off OpenCL's head IMHO.
 
Ryokurin said:
It's open sourced as in how microsoft does 'Shared Source' that is, you can apply and they may approve you access, but only enough to do what you need to do, not enough to completely figure out how it works or say port it to another card. This is more of a move to assist developers who want to make ARM + Nvidia clusters than a move to cut off OpenCL's head IMHO.
Click to expand...

ARM + nvidia clusters would make some sense if nvidia wants to continue to seperate itself in a market that is in it's infancy and looks like it may takeoff in a couple years. Not that nvidia has any choice with no x86 license.

Still think there is missing a piece of this puzzle.
 
You must log in or register to reply here.
Back
Top