(j3.2006) OpenCL usage compared to OMP, MPI, and coarrays
Tom Clune
Thomas.L.Clune
Tue Oct 4 13:41:14 EDT 2011
I was at a workshop for GPU's in climate/weather models in Boulder 2-3 weeks ago. OpenCL is not currently seen as a major player in that community. Presumably this mostly reflects the general Fortran bias in that community, but the success rate with CUDA for Fortran and even some impressive results with directives were the ultimate factors. John Leveque (Cray) made a strong case for the fact that directives should soon catch up with hand-tuned CUDA in many situations. PGI did not dispute the claim, and the various compiler types in the room did not see it as an outlandish statement. (Surprised me though given general experience with OpenMP performance.)
Cheers,
- Tom
On Oct 4, 2011, at 1:30 PM, Bill Long wrote:
> I'm guessing members of an OpenCL user's group. Or maybe their own
> customers. What I see, for "on node" threading is a LOT of OpenMP and a
> bit of CUDA. When the accelerator (i.e. GPU) directives go onto
> OpenMP, I don't see much of a reason for OpenCL to stay around.
>
> Cheers,
> Bill
>
>
> On 10/4/11 12:08 PM, Rasmussen, Craig E wrote:
>> Interesting results from AMD
>> http://developer.amd.com/afds/assets/keynotes/1-Bergman-FINAL.pdf. I
>> wonder whom they polled.
>>
>> -craig
>>
>
> --
> Bill Long longb at cray.com
> Fortran Technical Support & voice: 651-605-9024
> Bioinformatics Software Development fax: 651-605-9142
> Cray Inc./Cray Plaza, Suite 210/380 Jackson St./St. Paul, MN 55101
>
>
> _______________________________________________
> J3 mailing list
> J3 at j3-fortran.org
> http://j3-fortran.org/mailman/listinfo/j3
Thomas Clune, Chief
Software Systems Support Office
NASA GSFC (610.3) B33-C128
301-286-4635 (W)
<Thomas.L.Clune at nasa.gov>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://j3-fortran.org/pipermail/j3/attachments/20111004/77135fa1/attachment.html>
More information about the J3
mailing list