Re: is Vax addressing sane today

On Sun, 6 Oct 2024 23:42:47 -0000 (UTC)
Lawrence D'Oliveiro <ldo@nz.invalid> wrote:

On Sun, 06 Oct 2024 07:18:59 GMT, Anton Ertl wrote:
 

However, in most supercomputers (including Summit) the GPGPUs
provide the bulk of the FLOPS ... 

 
That tends to go back and forth, between CPU and GPU.
 
See this
<https://www.nextplatform.com/2020/03/05/software-evolution-on-ornls-summit-supercomputer/>
interview with Dr Tjerk Straatsma, group lead for scientific computing
at ORNL. Seems their supers have made heavy use of NVidia GPUs up to
now, but this was set to change:
 
    Frontier, the next system for the OLCF, will have AMD CPUs and
    GPUs.
 

No back and forth here. Frontier is as much GPU-centric as was Summit.
The same for Aurora that is installed alongside earlier Polarais at
Argonne.
Or for El Capitan that is going to replace Sierra at LLNL.

In all cases the vendor of GPU changed, but balance between GPU and CPU
computing power remains heavily skewed toward GPU.

    To prepare for this system, software developers may want to make
    changes to their programming approach, with OpenMP directive-based
    and HIP native offloading as the most comparable to the OpenACC
    and CUDA approaches on Summit today.
 
I wonder what happened to OpenCL as the cross-platform architecture
for GPU computing?

I have no idea.
May be, it was not designed with the same level of competence as CUDA ?
Or, may be, being cross-platform, OpenCL is at inherent disadvantage?