On Tue, Jan 17, 2017, at 14:11, Guillermo Rozas wrote: > Try increasing the opencl_memory_headroom parameter as mentioned by > Matthias before. It worked for my 960M for those particular images.
Thanks, I will, if I ever need to for my actual use case. Other than an occasional stitched panorama, I'm usually dealing with 16MP images, and so far haven't encountered any tiling issues with those (even when testing with the equalizer, which I rarely use for general purposes). This is another example of how simplistic benchmark numbers can be misleading: even if a person looking at them allowed for the fact that the test image was larger, and hence slower, than their own use case, they might very well assume a linear relationship, when the reality about tripping a threshold or not. It also seems like if you're comparing the cards to a particular CPU (i7 or what have you), you should really be using that CPU for the card tests, too... -- jys ____________________________________________________________________________ darktable user mailing list to unsubscribe send a mail to [email protected]
