<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Fri, Jul 14, 2017 at 10:22 AM, Karl Rupp <span dir="ltr"><<a href="mailto:rupp@iue.tuwien.ac.at" target="_blank">rupp@iue.tuwien.ac.at</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><span class=""><br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
it will nonetheless require a lot of convincing that at best they<br>
get moderate speed-ups, not the 580+x claimed in some of those early<br>
GPU papers...<br>
<br>
<br>
Karli, we are talking about two different things. You are talking about performance, and I applaud you for that, but I am talking about giving customers what they want. They want to investigate GPUs. I will say that I do not anticipate seeing any performance improvement.<br>
</blockquote>
<br></span>
it is fine to say that "here is what you can try, but don't expect performance gains". My experience is, however, that people then go ahead, try it, *and* expect performance gains (of course for inappropriate system sizes, etc.). Since they invested time in exploring GPUs, there is an implicit expectation that there must be a ROI...<br></blockquote><div><br></div><div>Yes. In the particular case that I am working with I have made it clear that they should expect performance gains. And the postdoc that is doing this is capable and reasonable and understands that this is not likely to be faster. If hypre works out of the box, with maybe some manual movement of data to the GPU, say, then he might do it just to have a bullet.</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
Best regards,<br>
Karli<br>
<br>
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><span class="">
<br>
<br>
The fusion folks that I work with, and I assume other DOE<br>
offices, are just looking at their codes, subroutine by<br>
subroutine, and having postdocs look at GPUising them. We just<br>
need intelligent answers to their questions. Even if we as<br>
sentient and passionate human being have opinions on the<br>
approach that is implied by their questions, it is part of my<br>
job to just give them a professional answer.<br>
<br>
<br>
In the past ~18 months I've worked with applications that wanted to<br>
use GPUs in just that manner. Needless to say that you end up with<br>
touching almost everything to actually beat an existing (efficient)<br>
CPU-based application by less than a factor of 2. This involves<br>
MPI-parallel applications; it's much easier to get higher speedups<br>
if you don't need to communicate across ranks.<br>
<br>
<br>
I have enough now (thanks Jed and Lorena, et al!) to answer the<br>
AMGx question sufficiently, and if you could give me a quick<br>
assessment of where we are with hypre's GPU solver that would be<br>
great.<br>
<br>
<br>
does "work in progress" suffice? ;-)<br>
<br>
<br>
I will advertise it "as is" (this is a term of art in US law).<br>
<br>
Thanks again,<br>
<br>
<br>
Best regards,<br>
Karli<br>
<br>
<br>
<br>
<br>
<br>
<br>
On Thu, Jul 13, 2017 at 11:16 PM, Karl Rupp<br>
<<a href="mailto:rupp@iue.tuwien.ac.at" target="_blank">rupp@iue.tuwien.ac.at</a> <mailto:<a href="mailto:rupp@iue.tuwien.ac.at" target="_blank">rupp@iue.tuwien.ac.at</a>><br></span>
<mailto:<a href="mailto:rupp@iue.tuwien.ac.at" target="_blank">rupp@iue.tuwien.ac.at</a> <mailto:<a href="mailto:rupp@iue.tuwien.ac.at" target="_blank">rupp@iue.tuwien.ac.at</a>><wbr>>><span class=""><br>
wrote:<br>
<br>
Hi Mark,<br>
<br>
I hear Hypre has support for GPUs in a May release. Any<br>
word on<br>
the status of using it in PETSc?<br>
<br>
<br>
as far as I know, it is currently not supported in PETSc.<br>
I'll have<br>
a look at it and see what needs to be done to enable it.<br>
<br>
<br>
And we discussed interfacing to AMGx, which is complicated<br>
(precluded?) by not releasing source. Anything on the<br>
potential<br>
of interfacing to AMGx? I think it would be great to<br>
make this<br>
available. It is on a lot of checkboxes. I would love<br>
to be able<br>
to say, yea you can use it.<br>
<br>
<br>
Lorena Barba's group actually interfaced PETSc to AMGx at<br>
some point<br>
(presented at GTC 2016 if I'm not mistaken). I'll reach out<br>
to them,<br>
maybe they have something to contribute.<br>
<br>
Best regards,<br>
Karli<br>
<br>
<br>
<br>
</span></blockquote>
</blockquote></div><br></div></div>