AMD Launches ‘Boltzmann Initiative’ to Dramatically Reduce Barriers to GPU Computing

Yes, yes, and yes... finally.

More developer tools and support. This is a good step, now AMD needs to keep building up on this.

Hilarious how they have to make a program to try and get people to port their HPC apps from CUDA.

Makes no difference if people do not use it. I have seen so many promises from AMD over the last few years, I do not believe them anymore until I see it.

AVogel said:

Hilarious how they have to make a program to try and get people to port their HPC apps from CUDA.

Click to expand...

It is perhaps of some value that the tool ports from CUDA to C++. As with some of their other products, they love to trumpet that nVidia is proprietary and they are open standards-based. Of course, I think they would have gone the proprietary route if they thought they could get away with it, but they don't have the market clout.

That begs the question, though, of how optimized CUDA code is to run at maximum efficiency and performance on Tesla/Quadro hardware vs. how bloated and slow C++ may be on AMD hardware, because of the lack of customization.

All of which leads us to the following excellent point:

ManofGod said:

Makes no difference if people do not use it. I have seen so many promises from AMD over the last few years, I do not believe them anymore until I see it.

Click to expand...

Even being open-source can't help you if you have no market traction. And being proprietary doesn't really hurt nVidia's customers--it's not like they have to pay for CUDA.

Commander Suzdal said:

It is perhaps of some value that the tool ports from CUDA to C++.

Click to expand...

I don't think you know anything about CUDA, especially because CUDA applications are written in C or C++.

The problem with AMD's many failed GPGPU initiatives are 1) AMD has abandoned most in favor of "something better", 2) no one writing HPC GPGPU code cares about it being multi-platform: optimizing it for one hardware platform is fine, so they completely ignore OpenCL in that field, 3) AMD does not provide the level of support or tools that Nvidia does (first class libraries and compilers, plus great 3rd party support) and instead primarily relies on contributions to open source projects, 4) AMD has crappy documentation and support for GPGPU, which leaves only crazed partisans not understanding why AMD is out in the cold for GPGPU.

Funny headline from *2* years ago (November 14th 2013):

"NVIDIA Dramatically Simplifies Parallel Programming With CUDA 6"

The big features on that release were unified CPU/GPU memory support and automatic BLAS and FFTW GPU optimizations for CPU code (drop in replace library and automagically get "up to 8x speed improvement" with no source code changes), supported on both single and multi-GPU configurations.

pxc said:

Funny headline from *2* years ago (November 14th 2013):

Click to expand...

I thought you were going to quote Roy Taylor when he said CUDA was doomed because nobody in the industry liked it.

pxc said:

I don't think you know anything about CUDA, especially because CUDA applications are written in C or C++.

Click to expand...

Well, no, I don't know anything about CUDA except a rough outline of what it's used for, but my comment was based entirely on the quote from the article. If that quote is off the mark, perhaps address your thoughts there?

Commander Suzdal said:

It is perhaps of some value that the tool ports from CUDA to C++. As with some of their other products, they love to trumpet that nVidia is proprietary and they are open standards-based. Of course, I think they would have gone the proprietary route if they thought they could get away with it, but they don't have the market clout.

Click to expand...

Except there was a time when AMD did have the market clout and they still didn't do what Nvidia has done. Nvidia was doing proprietary even when they weren't the market leader, they have just been getting worse since they got it.