> On Feb 11, 2021, at 1:56 PM, Atchley, Scott wrote:
>
>> On Feb 11, 2021, at 1:11 PM, Jeff Squyres (jsquyres) via devel
>> wrote:
>>
>>
>>
>> That being said, we just added the AVX MPI_Op component -- equivalent
>> components could be added
> On Feb 11, 2021, at 1:11 PM, Jeff Squyres (jsquyres) via devel
> wrote:
>
>
>
> That being said, we just added the AVX MPI_Op component -- equivalent
> components could be added for CUDA and/or AMD's GPU (what API does it use --
> OpenCL?).
AMD’s API is HIP:
https://rocmdocs.amd.com/en
> On Feb 28, 2018, at 2:12 AM, Pavel Emelyanov wrote:
>
> On 02/27/2018 05:18 AM, Dmitry V. Levin wrote:
>> On Mon, Feb 26, 2018 at 12:02:25PM +0300, Pavel Emelyanov wrote:
>>> On 02/21/2018 03:44 AM, Andrew Morton wrote:
On Tue, 9 Jan 2018 08:30:49 +0200 Mike Rapoport
wrote:
>
I, with one task per thread
> - Hybrid MPI+OpenMP, a task is bound to a core or socket, but never to a
> thread
>
> Cheers,
>
> Gilles
>
> On Thursday, March 24, 2016, Atchley, Scott wrote:
> Hi Aurélien,
>
> I have said the same thing to many users over the years. Our c
Hi Aurélien,
I have said the same thing to many users over the years. Our colleagues at
NERSC, however, have found that 20% of their codes work better when using HT.
Some codes benefit from SMT2 (i.e. HT) and even SMT4 (available on Power8) in
order to provide enough latency hiding of memory ac
On Oct 21, 2015, at 11:09 AM, Jeff Squyres (jsquyres)
wrote:
> REVISION 2 (based on feedback in last 24 hours).
>
> Changes:
>
> - NETWORK instead of NETWORK_TYPE
> - Shared memory and process loopback are not affected by this CLI
> - Change the OPAL API usage.
>
> I actually like points 1-8
ers)
>
> Cheers,
>
> Gilles
>
>
> On Wednesday, October 21, 2015, Atchley, Scott wrote:
> On Oct 20, 2015, at 4:45 PM, Jeff Squyres (jsquyres)
> wrote:
>
> > On Oct 20, 2015, at 3:42 PM, Jeff Squyres (jsquyres)
> > wrote:
> >>
> >>
On Oct 20, 2015, at 4:45 PM, Jeff Squyres (jsquyres) wrote:
> On Oct 20, 2015, at 3:42 PM, Jeff Squyres (jsquyres)
> wrote:
>>
>> I'm guessing we'll talk about this at the Feb dev meeting, but we need to
>> think about this a bit before hand. Here's a little more fuel for the fire:
>> let's
What is the output of /sbin/lspci -tv?
On Aug 31, 2015, at 4:06 PM, Rolf vandeVaart wrote:
> There was a problem reported on the User's list about Open MPI always picking
> one Mellanox card when they were two in the machine.
>
> http://www.open-mpi.org/community/lists/users/2015/08/27507.php
This header file’s macro name does not match OMPI style.
Typically, the macro name is FRAMEWORK_COMPONENT_H. Note, the #endif at the
end of the file is correct (i.e. OPAL_DL_H).
Signed-off-by: Scott Atchley
diff --git a/opal/mca/dl/dl.h b/opal/mca/dl/dl.h
index 7c6d2ef..1e92c1a 100644
--- a/op
We have some new Power8 nodes with dual-port FDR HCAs. I have not tested
same-node Verbs throughput. Using Linux’s Cross Memory Attach (CMA), I can get
30 GB/s for 2 MB messages between two cores and then it drops off to ~12 GB/s.
The PCIe Gen3 x16 slots should max at ~15 GB/s. I agree that when
Hi George,
The Power8 can run in little-endian mode without penalty:
http://www.hpcwire.com/2014/04/23/power8-openpower-might-mean-hpc/
Not saying hetero support is unneeded, but this case may not be it.
Scott
On Apr 24, 2014, at 12:54 PM, George Bosilca wrote:
> There seems to be an opportu
12 matches
Mail list logo
