All joking aside - the best benchmark is your code. I know both hardware so intimately well that I could easily show codes which perform well on both sides. If you are looking to build a system and need specific advise; I'm happy to work with you offlist.
My original point was that it's unfair to quote single precision numbers and say 2x when double precision or half precision performance is not a similar story. Architecturally there are also different - AMD dGPU may be more forgiving for codes which have higher register pressure and they implement a native IEEE 64bit fp div and NVIDIA cards may have an advantage in other categories.. it all depends. On Fri, Jul 29, 2016 at 12:40 AM, Michael Di Domenico <[email protected]> wrote: > On Thu, Jul 28, 2016 at 12:23 PM, C Bergström <[email protected]> > wrote: >> On Fri, Jul 29, 2016 at 12:17 AM, Michael Di Domenico >> <[email protected]> wrote: >>> On Thu, Jul 28, 2016 at 12:06 PM, C Bergström <[email protected]> >>> wrote: >>>> Dum dee doo.... I welcome hard real world numbers from anyone who can >>>> produce them. >>> >>> what would you proclaim as a definitive application to run, for the >>> list to use as real-world hard numbers? :) >> >> I'll show you mine if you show me yours ;) > > don't have any to show, hence my prodding... i'd be happy to run > tests on the smattering of hardware i have at my disposal and provide > numbers to the list, but we should come to some consensus on the app > and how the numbers are interpreted, first. i think others would be > willing to do the same... > _______________________________________________ > Beowulf mailing list, [email protected] sponsored by Penguin Computing > To change your subscription (digest mode or unsubscribe) visit > http://www.beowulf.org/mailman/listinfo/beowulf _______________________________________________ Beowulf mailing list, [email protected] sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
