One question, looking at the image (screenshot of the processing menu), would it be advisable to disable the storage of previews and thumbnails so that the program runs more smoothly?
Another question. Working on Windows, where are the caches stored?
Thanks a lot for the new release! I’m a devoted DT user since v0.6!
However, I’ve noticed few things to improve.
Couldn’t find how to set sigmoid to default instead of filmic. I expected such an option in preferences menu but apparently it’s not there.
The balanced “OpenCL vs CPU” tiling went wrong. If darktable resource = large and OpenCL scheduling profile = default are set it estimates GPU time ~95s CPU ~5.5s but in fact it takes about 10 min (x100 vs estimation) istead! Setting darktable resource = unrestricted and OpenCL scheduling profile = very fast GPU disables this mode so works as a workaround to me speeding up export with upscaling x5! (16G RAM 8GB Radeon Ellesmere GPU).
Anyway, DT is really getting better and better with every release, so thanks again!
We’re kinda moving away from using preference options to auto-enable things. If you want to disable filmic and enable sigmoid you should set your workflow (preferences > processing > auto-apply pixel workflow defaults) to “none” and then set up an automatically-applying preset to enable sigmoid.
I couldn’t find how to enable/disable it. However, it was obviously on. In the first place I noticed huge time to export photo, so I started dt with -d opencl -d perf and looked into the log (unfortunately I did not tee’d the output so it’s lost). I saw the statement that dt estimates CPU to be more effective so it was chosen for processing.
Now I reversed the changes and am trying to make it use tiling to reproduce the previous state and store the log into file. But it doesn’t use tiling anymore… weird…
Haven’t touched it… Currently it reads as cldevice_v4_gfx803=1 150 0 16 16 1024 1 0 0.033319 0.000
if I understand correctly what you meant.
l look at numbers in lines such as 1083.313032 [dev_pixelpipe] took 65.222 secs (3.772 CPU) [export] processed diffuse or sharpen LC’ on GPU with tiling, blended on CPU`
However, I did not measure anything. When I looked into the log, there was a line saying that DT had estimated GPU vs CPU and chose CPU for processing (sorry I can’t recall exact wording). And some numbers were given in this statement as estimated processed times.
That’s okey. Will have a look there… or just wait.
UPD: That was the first run of 4.2. May be some surplus or whatever…
Ok, but why, though?
It would make sense to me that I could easily set my preference for this since it is a basic workflow and I have a choice between two equally good options.
Setting up styles seems somewhat roundabout.
Anyway, thanks for giving me all these great options to choose from
Well, I think “avoid atomics” and “micro nap” most probably are a surplus from HD7770/fglrx times. So I did touch these two a decade ago or so… However I doubt these are involved. I may reset them to defaults, though.
Yeah, I agree that sigmoid as default option is a good choice. When someone needs more control in difficult cases one may switch to filmic. So maybe the opportunity to quickly switch by a shortcut or a mouse click will be convenient.