Okay - so then we should probably skip this in the release version - or at least keep it as a developer only option for the release since I can see it causing issues with people using that option and then complaining that their executable does not work on other hardware. Keep in mind also the average user doesn't know what hardware class their GPUs are.
Leaving off the -volta option will result in an executable that works on Kepler, Maxwell, Pascal and Volta yes?
All the best
Ross
> On Mar 24, 2018, at 23:39, David Cerutti <dscerutti.gmail.com> wrote:
>
> The optimization with -volta is about 10%. All of this is soon to change,
> anyway, I hope, and there is precious little volta hardware out there at
> the moment--it's probably mostly Titan-V, which I haven't even tested on
> myself yet.
>
> Dave
>
>
> On Sat, Mar 24, 2018 at 10:08 PM, Scott Brozell <sbrozell.rci.rutgers.edu>
> wrote:
>
>> Hi,
>>
>> IMO, the highest priority for others tackling 3 and 4 is to build
>> with the latest intel compilers (which looks like 18.0.2).
>> More detailed comments below.
>>
>> On Sat, Mar 24, 2018 at 10:38:34AM -0400, David A Case wrote:
>>> 3. Does anyone besides Scott have access to Intel 18.0 compilers?
>>> Basic things like the dhfr or water tests seem to be failing there. If
>>> this is reproduced, can someone put a trap in the configure2 script?
>>
>> Note that 18.0.2 is the latest (i used 18.0.0). My guess on the 18.0.0
>> major failures is a change in optimization policy on Intel's part.
>> We have been using 18.0.0 for a couple of months at OSC w/o problems,
>> eg, quantum espresso and other sci. apps. But I'm skeptical of x.0.0
>> versions in general.
>>
>> I plan to test with other older versions.
>>
>>> 4. The PGI compilers are favorites for some people, and we need a
>>> second opinion here, vs the srb results on the wiki. If anyone has
>>> access to other versions, that might be worth a shot as well.
>>
>> The pgi 17.10.0 results which include a number of seg faults are
>> probably par. Whether lurking bugs in our code or thiers, pgi has
>> frequently been more touchy. It might be prudent to mention these
>> issues in a README, KNOWN_FAILURES, or the like.
>>
>> I plan to test with other older versions.
>>
>> scott
>>
>>
>> _______________________________________________
>> AMBER-Developers mailing list
>> AMBER-Developers.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber-developers
>>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sun Mar 25 2018 - 10:00:03 PDT
