[gmx-developers] GROMACS OpenCL on Gallium

[Szilárd Páll]

On Thu, Nov 26, 2015 at 8:21 PM, Vedran Miletić <rivanvx at gmail.com> wrote:

> 2015-11-26 20:07 GMT+01:00 Szilárd Páll <pall.szilard at gmail.com>:
> > Hi,
> >
> > Besides FPGA folks, what about Apple, embedded and mobile platforms
> > (Qualcomm, ARM, Samsung, etc.)?
> >
> > I'm not sure Intel is totally uninterested. They've just moved out again
> > their OpenCL SDK form the silly bundle they had before in the latest
> release
> > AFAIK because people complained.
> >
>
> On a non-HPC side, there is also Beignet which, from what I hear,
> should be pretty stable.
>
> > NVIDIA: no comment.
> >
>
> Actually, they have improved in this regard. They now support OpenCL
> 1.2 [1]


Yeah. I'm still not buying that they don't have enough request for OpenCL
support to put a guy or two on it and
i) implement standards in a timely fashion
ii) fix bugs that render their runtime useless for most intents and
purposes (e.g. the blocking clEnqueueNDRangeKernel, a ~6 years old bug that
we have observed recently too [1] is just insane).

And let me quote someone else with controversial claims:
https://twitter.com/jrprice89/status/667466444355993600



> and they are willing to fund OpenCL-based teaching and
> research programs; they have rebranded their academic programs from
> CUDA Teaching Center to GPU Education Center etc.
>

I did not know about it, kudos for that.


>
> > HIP and CUDA support seems like a desperate move from AMD to lower the
> > barrier of entry and make things (seem) easier. Attracting dev/user
> interest
> > to stay afloat is crucial for them. It would however be a major mistake
> for
> > AMD to move away from OpenCL, I think - unless they want to shoot
> themselves
> > in the foot by encouraging people to only write CUDA kernels for AMD.
> Still
> > need to look into this closer to understand what the direction is.
> >
>
> I agree completely. I hope they are aware that being completely open
> source is their (only?) big advantage, because stability, performance,
> and performance per watt are lacking.
>

I mostly agree. Open source could be a way to gain traction and advantage.

However, there was a window for AMD to gain some traction - sadly past
tense as now it's becoming too late unless Pascal slips and miracles
happen. Their raw DP (and in some cases SP) flop rate is better on the
latest FirePro GPUs and the new Maxwell surprise Teslas can't compete with
that. COmbine that with 32 GB/board, is something that NVIDIA does not have.


> > Overall, I feel like this is the time to not take the back seat. Rather
> than
> > letting others decide whether it's going to be open standards or vendor
> > lock-in that defines the low-level accelerator programming for the coming
> > years I feel like we, though GROMACS, can show that we care and perhaps
> can
> > make a difference. That's why I wrote the previous mail. Don't get me
> wrong,
> > I do not have the illusion that tomorrow we can just drop CUDA support
> just
> > to make a point. However, providing an a decent alternative based on
> OpenCL
> > and pointing out that we want the open alternative to work as well as the
> > closed one does require effort, but it is realistic.
> >
>
> Thanks for saying that. GROMACS is an widely used code and we might
> have a chance here.
>
>
[1.]
https://forums.khronos.org/showthread.php/5822-non-blocking-call-to-clEnqueueNDRangeKernel

Cheers,
--
Szilárd




> Regards,
> Vedran Miletic
>
> [1]
> http://www.geeks3d.com/20150326/nvidia-r349-12-graphics-driver-for-linux/
>
>
> > Cheers,
> >
> > --
> > Szilárd
>
>
> --
> Vedran Miletić
> http://vedranmileti.ch/
> --
> Gromacs Developers mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
> or send a mail to gmx-developers-request at gromacs.org.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20151126/31f1541f/attachment.html>