AMD To Support CUDA

[Smokingenius]

I thought this was an interesting release today. Just thought I would share it.

http://www.anandtech.com/show/9792/a...s-for-amd-gpus

AMD is announcing they will be releasing compilers for their cards that run C++ and Cuda. This is great news for us AMD fans in general, as we can take advantage of CUDA with AMDs new HBM memory architecture. Should make parallel processing more interesting. Combine that with Intel's Knights Landing supercomputer on a chip, and we got a lot of number parallel processing opportunities coming down the pipe.

[Mark Rose]

Interesting!

[jwaltos]

This contained link is interesting also: http://www.phoronix.com/scan.php?pag...UDA-GPGPU-Comp

[TheJudger]

Sounds too good to be true to me.
I would really be surprised if this works for (almost) all codes and runs at reasonable speeds.

Oliver

[xilman]

Originally Posted by TheJudger

Sounds too good to be true to me.
I would really be surprised if this works for (almost) all codes and runs at reasonable speeds.

Oliver

Maybe.

Maybe if this announcement had been made a month ago I would have considered something other than a GTX-970 for my new machine.

[chalsall]

Originally Posted by TheJudger

Sounds too good to be true to me.
I would really be surprised if this works for (almost) all codes and runs at reasonable speeds.

Oliver

That's the catch...

Emulation is easy. Speed, not so much....

[TObject]

LOL

Just in time for the Microsoft’s announcement that they are giving up on running Android applications under Windows 10.

[ewmayer]

Originally Posted by TheJudger

Sounds too good to be true to me.
I would really be surprised if this works for (almost) all codes and runs at reasonable speeds.

Oliver

My initial optimism at the headline mostly evaporated by the time I finished reading the piece - besides the translated-code performance angle, based on the last few paragraphs I am somewhat pessimistic that AMD will be able to do the CUDA translator as they say they plan to without getting sued by nVidia. We shall see. Rest of it confirms my take w.r.to AMD having failed to get significant traction with their GPU/toolchain offerings.

[chalsall]

Originally Posted by ewmayer

My initial optimism at the headline mostly evaporated by the time I finished reading the piece - besides the translated-code performance angle, based on the last few paragraphs I am somewhat pessimistic that AMD will be able to do the CUDA translator as they say they plan to without getting sued by nVidia.

IIUC, AMD and NVidia have very different approaches to parallelism. This is why AMD is better at BitCoin mining than NVidia.

Emulation of compute is always possible; efficiency is quite another dimension to the equation. What's more expensive, the hardware or the human over the projected deployment time?

[kladner]

Originally Posted by chalsall

IIUC, AMD and NVidia have very different approaches to parallelism. This is why AMD is better at BitCoin mining than NVidia.
Emulation of compute is always possible; efficiency is quite another dimension to the equation. What's more expensive, the hardware or the human over the projected deployment time?

This made me wonder about parallelism. I am still wading through the first search result I picked to read. I can get an idea of the broad outlines, at least. I am not a coder, though I do try to study the examples
The article is
SIMD < SIMT < SMT: parallelism in NVIDIA GPUs
November 10th, 2011

My search was 'amd vs nvidia parallelism' I think some of the other hits may address either AMD parallelism, or some kind of comparison.
In fact:
Architecture Comparisons between Nvidia and ATI GPUs: Computation Parallelism
and Data Communications -from LSU.EDU pdf GTX 580 and HD 5870 era.

Understanding the parallelism of GPUs
-from
Rendering Pipeline
from geometry to pixels