Hi Jochen thanks for the example files. I ran it with my benchmark script, 47sec on the Radeon Pro, 60 secs on the Intel Iris and 66secs on CPU only... looks like your xmp is using modules that are CPU bound. On my regular benchmark egpu is 3x faster than intel gpu and egpu is 7x faster than cpu-only - not the case with your xmp file where egpu isn't even twice as fast vs cpu-only. comparing your log (first line) to mine (second):
53,785899 [dev_process_export] pixel pipeline processing took 53,032 secs (130,516 CPU) 48.591362 [dev_process_export] pixel pipeline processing took 47.065 secs (206.542 CPU) you can see that a lot of time is spent on cpu, so get more cpu cores instead of a new gfx card :) No just joking, maybe our gpu cannot do some stuff that makes it run on cpu? Or the module cannot run on opencl ever due to gpu limitations? Somebody more educated on the opencl code might have an answer. And after that answer go shopping. hth mike Am Mo., 16. Dez. 2019 um 21:03 Uhr schrieb Jochen Keil < [email protected]>: > Hi Michael, > > thanks for your reply! I'm pretty confident that darktable is using GPU > acceleration. I've checked the output of `darktable-cltest` several times > (like someone else suggested) and it says that OpenCL is available and in > use. > > For reference, here's a RAW + XMP: > > https://drive.google.com/open?id=1FBunVv3reyUTXIXFFGP2dE02EAyHvkoa > https://drive.google.com/open?id=1LEs9cIo760yplkyE57S_5Mqkm8-pVATl > > Attached you'll find the log of > > `$ time darktable-cli 2019-11-23T23:23:35+0100_7871.arw > 2019-11-23T23:23:35+0100_7871.arw.xmp 2019-11-23T23:23:35+0100_7871.png > --core --library :memory: -d opencl -d perf 2>&1 > dt_log.txt` > > I hope this will shed some light on the issue. Or maybe I just went (too) > crazy on the modules :) > > Thanks and best, > > Jochen > > > On Mon, Dec 16, 2019 at 2:27 PM Michael Kefeder <[email protected]> > wrote: > >> Hi Jochen, >> >> Have you confirmed darktable is actually running GPU acceleration? e.g. >> looking at the logs of darktable (e.g. using arguments: -d opencl -d perf)? >> I ran benchmarks on my computer with a 121MB RAW file (download of a Fuji >> GFX100 DNG) with 17 modules active and that takes 34sec to export, on an >> eGPU "AMD Radeon Pro 580 Compute Engine 8192 MB" which is performance wise >> in the ballpark region of your card. egpu attached to a MacBook Pro, 2.7 >> GHz Intel Core i7, 16GB ram. Since I doubt you have more than 17 modules >> active and a faster desktop machine I suspect your 1 minute time is caused >> by something else than your GPU. >> >> hth >> mike >> >> Am Mo., 16. Dez. 2019 um 11:51 Uhr schrieb Jochen Keil < >> [email protected]>: >> >>> Dear all, >>> >>> two years ago I bought a "Palit GTX1060 Super Jetstream 6GB. PCIE 3.0 / >>> 6GB DDR5 192bit 8.0GHz / 1847MHz / 1280 Cuda Cores" specifically for the >>> task of editing pictures in darktable. >>> >>> In the meantime I upgraded various other components (RAM, CPU) but >>> overall, it still takes about a minute to export a picture (~80MB RAW -> >>> 16bit TIFF). Additionally, editing can get pretty slow with many modules >>> applied. >>> >>> Will upgrading to a newer GFX give me a significant performance boost? >>> Any GFX recommendations for that purpose? >>> >>> Thank you very much and best wishes, >>> >>> Jochen >>> >>> ____________________________________________________________________________ >>> darktable user mailing list to unsubscribe send a mail to >>> [email protected] >>> >> ____________________________________________________________________________ darktable user mailing list to unsubscribe send a mail to [email protected]
