hi, i suppose running darktable-cli with --core -d perf will give some insight into whether or not the cards needed to resort to tiling to fit the buffers into memory. that's a huge speed impact (multiple passes necessary + process the overlap more than once) and can depend on many things (actually available memory on the card, how much can be allocated by the driver and how much is in use by others, potentially).
-jo On Fri, Dec 2, 2016 at 11:02 AM, Patrick Rudin <[email protected]> wrote: > Denis Testemale <[email protected]> wrote: > >> I gave a shot at this test, out of curiosity, and can't really >> understand the results because they seem much better than the ones >> published (especially test 2 and 3), > > Well, number 1 are the modules equalizer and denoise. And i did it with > darktable 2.0.7. > > Michael created tests 2 and 3 himself, only god knows which modules he > used. But he used darktable 2.2.x do create the xmp, so i guess there > are some (not visible) errors if you use it with darktable 2.0 > > Run the "benchmark" with darktable-cli, and you will see the full > output. > > Personally i don't mind if some card is 0.3 seconds faster than another > card. I guess it is interesting for many people to see which > card/driver-combination is working. > > So far, i have seen good results from nvidia 760/1060 and higher as > with radeon 370x/rx460 and higher. What about older and cheaper cards, > what about passive cooled cards, do they bring any advantage over cpu? > > Regards > > Patrick > ____________________________________________________________________________ > darktable user mailing list > to unsubscribe send a mail to [email protected] > ____________________________________________________________________________ darktable user mailing list to unsubscribe send a mail to [email protected]
