Are there any power settings you need to configure esp with a laptop. In windows globally you would set the high or ultra high performance profile so that things don’t throttle down and then on an app by app basis you can also designate the app as high performance and to use the GPU… not the best maybe for battery but that should guarantee max performance on the OS side… I have no idea about Linux laptops but there might be something similar???
Gnome DE had power modes but I think they never really worked. Then there were (I think?) Gnome extensions that let you define your own power modes and select them via a drop-down menu, but that just limited the CPU max frequency as far as I know.
It meant that if something was processed at 20% CPU at 3GHz, then limit to half the frequency (1.5GHz) would double the CPU usage (40%), so in the end it would balance itself out and no power was saved…
The best way to save power was actually using the cpufrequtils package (I think the program itself was called auto cpu-freq) and set power governor settings of my CPU which auto switched to powersave on battery and performance on plugged in. I have a modern Intel processor, so it might not work for everyone.
I’m not aware of any app-specific settings through, you’ll need to try and look that up.
good, this is the better one of the two models mentioned on the website…
1 Like
+1 darktable, works fine for me with a 4k screen. It’s all about the hardware in the end.
2 Likes
You guys remind me of my deceased father:
Many years ago my cousin’s son was little and I knitted a pullover for him with a very complex Norway pattern. Unfortunately the sleeves of the pullover were too long and there was no way to make them shorter because of the pattern. My dad said: not the sleeves are too long but the child’s arms art too short. Of course he meant it as a joke.
5 Likes
Btw, there is no problem with the preview updating when there are just the few standard modules like sigmoid, color balance rgb, diffuse/sharpen etc. The speed problem occurs when I have very complex edits with 3 instances of color balance rgb with masks, several instances of exposure, tone equalizer etc. Which happens quite often in my case.
1 Like
Yes, there’s a strong streak of Stockholm Syndrome running in the entire FOSS community, and photo development in particular.
In some regards, Lightroom and Capture One and DxO Photo Lab et al have clear advantages. Unquestionably better denoising in Lr and DxO. AI masking in C1 and Lr. The processing speed and workflow streamlining of C1 and Lr. The overwhelming availability of great tutorials and edutainment available for Lr.
It is a bit strange how some part of the FOSS community seems to look down on commercial software with disdain. I suspect that for many folks, Lightroom on Windows or MacOS might actually be a better fit than stubbornly running Darktable on Linux in the name of some nebulous “freedom”. Being a programmer in my day job does highlight the inherent value of charging money for software.
But I’ve been there, I’ve used Lightroom and Capture One for several years, and for a multitude of reasons decided that darktable is a better fit for me. It is, by far, the most customizable of these tools, and through painful experience, I have learned that that’s of prime value to me. I probably really should write my own raw developer. Perhaps when I retire, or the kids have moved out…
1 Like
Sorry but having complete control of my stack is not nebelous at all.
At the OS level, windows will force you to install updates, and macOS will notify you constantly to do the updates. A lot of these updates are bringing features that are not useful for me as a user.
On the app level, I update DT when I’m ready. DT won’t rug pull me by changing to a subscription model, DT doesn’t have analytics, DT doesn’t phone home to tell the overlords what I’m doing. DT doesn’t stop working if I haven’t phoned home often enough. DT won’t ban entire countries from using it.
Not at all “nebulous” freedom. You speak from a provlidged position.
4 Likes
I didn’t mean it as if darktable is blameless, which it isn’t, or that it couldn’t be much more optimized, only that for my use case it works.
I also run heavy modules like 3 instances of D&S at 4k. I have an RTX3080 which doesn’t make this comparison very fair but still. Would you be up to sharing here a raw + sidecar with your workflow, so we could benchmark and check the differences?
I am also a programmer for a living and in my experience proprietary software is often worse and slower than the alternatives. Lately I’ve been forced to use Win for a few work related tasks and good lord have mercy on my soul. Good proprietary software exists, a programming example is JetBrains software collection, which is often pretty good and much better than other paid alternatives or FOSS alternatives.I still use nvim because I’m used to it and it fits my needs, but if I needed more powerful tools I would use something from JetBrains.
I also paid for Bitwig(flatpaked 
), which is proprietary and imo much better than Ardour or Reaper.
2 Likes
After what we’ve seen recently with forced Recall and Co-Pilot, even ignoring the entire history of proprietary OS’s spying on you, how can OS freedom be called nebulous?
1 Like
For me, photography is a hobby that takes up less than 10% of the time I spend in front of a computer. I am using Linux because it is rock-solid, stable, lets me deal with updates when I have the time, and integrates seamlessly into my workflow (write computational code on my laptop, send to a server for numerical computing, get the results back, analyze, repeat), which would be orders of magnitude more painful with Windows. I have my machines set up with continuous incremental backup, so if one of them croaks, I can get back my exact same environment within an hour, provided that I have spare hardware available.
Given that I am running Linux and shoot raw , Darktable, Art, and Rawtherapee are pretty much the only game in town. It’s not that I am stubbornly sticking to FOSS — if there was a paid alternative on Linux, I would consider it (but it would have to be really good to compete with the above). But there isn’t, because from the perspective of the relevant companies, I am a niche user. That’s fine, they ignore me, I ignore them back.
Possibly. But there is no way around the fact that pushing a high-resolution image through a pipeline of computationally complex algorithms is costly, so if you insist on 4k preview you need powerful hardware.
Your 4k laptop was probably designed for games, which are an order of magnitude less costly than the kind of calculations DT does. I doubt that LR on Windows/OS X is orders of magnitude faster for the same kind of algorithms (if applicable).
4 Likes
No, it was specifically designed for content creators or (media) work, a mobile workstation while not being too heavy (“studio” is part of it’sname), a gaming laptop should have a better gpu but a worse screen. A gaming laptop is probably better suited for darktable specifically.
2 Likes
Games are probably an order of magnitude more costly than what darktable does. Besides the fact that most run post processing algorithms similar to some implemented in darktable, they have to do it for every frame, multiple times a second. While also working on geometry, light, shadows, loading/unloading textures constantly, some games also use gpu’s for world physics calculations, etc. Sure they do this at lower resolutions, but need to output many more frames a second than a single photo.
A good example of how slow darktable is, is how fast vkdt is at running the same algorithms.
4 Likes
Yes, darktable processes the pixelpipes and that’s it, no processing done unless you interact with darktable again. Waiting a bit longer for dt to process something you did is a lot less painful than low fps in a game (for me at least).
1 Like
thanks. i feel like i must sound like a broken record repeating this all the time
darktable was designed 2009, and pure GPU computation was just not there yet. it had such severe limitations that it was simply not possible to deliver as software (no vulkan! limited texture size! no float16! no integers! not enough memory!). times have changed, and dt’s pipeline jumps some really wild hoops to make processing work somehow on limited hardware of the time (various async pipelines on fractions of the image data with wacky sync points between them and the gui etc). this means that simply stripping away this complexity removes so much performance loss between the cracks that today it’s easily possible to run much faster. that’s on today’s hardware though (or say 5 years old or maybe high end desktops from 10 years ago).
and yes, waiting for a few hundred milliseconds is painful but possible in the context of still photography. pretty much rules out raw video though. also faster processing means that you’ll be able to do more compute under your pain threshold. this might include some expensive neural network things for say denoising or masking. in some initial experiments i get about the dt ballpark of runtimes (50-100ms for network evaluation), but on the full res image instead of parts/downscaled previews.
6 Likes
I should have been more clear: I meant the cost of one “frame”. 120fps/4k is not uncommon on gaming PCs these days (of course depends on the game), while a 10fps pipeline would provide an entirely satisfactory user experience for photo editing.
Indeed it is remarkable, but my understanding is that vkdt is not running the exact same algorithms (some are not implemented yet). I am not sure if this is a technical constraint, or just lack of developer time though.
That said, I wonder if pure CPU performance can be improved in Darktable. SIMD, especially with recent advances, can be really fast (not as fast as a GPU though), but it is not always easy to get SIMD code out of LLVM, and it can be really fragile (in the sense that minor modifications can break it). AMD XDNA sounds really promising too, but I have not had time to look into the details.
(sorry for being off-topic)
2 Likes
I think stuff like demosaic, unsharp mask and other “base” algos are already implemented in both programs. Would be interesting to compare the performance between the two on the same system
1 Like
i am offended.
you know that the darktable code was traditionally really heavy on SIMD intrinsics, SSE4.2 at the time? and that there were efforts to replace this by something more generic, that would more naturally extend to AVX (openmp simd pragmas), with some medium success. to the contrary, for ARM compatibility it was necessary to remove the explicit SSE intrinsics again (so now dt has three codepaths for the same: x86/SSE, i386, opencl).
also, i recommend implementing a few algorithms using manual sse/avx intrinsics and then doing the same thing again using ispc or glsl, to develop a sense of preference for programming paradigm.
here’s a list of current vkdt modules. some are just a direct copy of the same algorithm as in dt, such as usm, llap, demosaic in RCD mode. negative, zones, filmcurv to some extent. some are adjusted because why would you make the same mistake twice. for instance denoise runs before black point subtraction/clipping at 0 so gaussian noise can average to black (not purple). that also means it runs on bayer data, so the input is different (changes memory bandwidth). i don’t intend to re-implement all dt modules, and certainly not for compatibility.
6 Likes
btw, I think it’s specifically the color equalizer that’s slowing down my preview, in dt 4.8 it uses the cpu not the gpu
2 Likes
I think its now built for opencl in the master …
1 Like