CUDA daddy muses on future GPUs

If you think of Nvidia as a hardware company, you're only three-quarters correct — at least according to Ian Buck, Nvidia's senior director of GPU computing software, who sees an even greater role for software as his company's products evolve. "It's important to note that software and hardware really are one," Buck told …

[Simon Brady]

Deja vu all over again

http://www.catb.org/~esr/jargon/html/W/wheel-of-reincarnation.html

But who needs history when you can have hype?

[Goat Jam]

I read it up until he said

"innovating"

I for one am totally sick of corporate mouth pieces using the "i" word almost as much as I am sick of politicians wheeling out "hard working families" at every opportunity.

[DrXym]

Dump CUDA

OpenCL is a platform neutral way of GPGPU programming. It's supported by a raft of industry players including AMD & NVidia (who were instrumental in developing it). Why are NVidia still pushing CUDA?

[Ru]

Re: Dump CUDA

Because it is widely used, well established, and reasonably engineered? Proprietary systems are unfortunate, but to throw it away merely because it is proprietary seems a little shortsighted, no?

[DrXym]

Deprecation

NVidia could easily say that version X is the last major release and then it goes into maintenance / deprecation mode. People can choose to stick with it in its final form or port to OpenCL.

I don't see this as being any different from (for example) Microsoft and Winforms, or Sun & AWT. Yes they'll give it token support but all the good stuff will go into another API which happens to be the industry standard.

[asdf]

easy performance

OpenCL performance is notoriously bad on Nvidia GPUs IIRC. Kind of defeats the purpose of using GPGPU farms for calcs if you are going to leave performance on the floor.

[Anonymous Coward]

is loving...

... the use of "munging", though I couldn't help fall back to my valley boy roots and read as is colloquially used.

Munging (v): equivalent to minging; abhorrent, disgusting, ugly

[RegReaderrr]

CUDA is Nice

As a person who wrestles with both CUDA and OpenCL I prefer CUDA by far. One can do a lot more in CUDA. OCL is basically meant for shaders, it's not the real deal. CUDA tools are excellent. No one comes close to that on the GPU side.

Give credit where credit is due, Nvidia is the ONLY company which is innovating in the area of processors. x86 is an ancient kludge, RISC has been around for a while. Only NVidia's Fermi is something new and it's quite fast and useful. In many fields it's the only game in town. The oil industry is a good example.