My current system is a i5-7600K paired with 32 GB RAM and an AMD RX 580 running arch linux.
What would you upgrade to make darktable processing faster. It’s bearable with my R5 files but I feel it could be a lot faster with more modern components. Yet prices are still a bit on the high side so I am seeking advice. I read about FP32 operations and many cards have many times more TFLOPS than the RX 580 but I can’t really decide which one to get. Also it should not be too power hungry but I fear every better card is more power hungry.
As a matter of interest, what are the computer bottle necks that slow down DT? Is it RAM, CPU, graphics card or what. I am running a computer that I don’t remember when I bought it but it was loaded with Windows 7 and I got a free upgrade to Windows 10 when it became available. I am satisfied with DT performance on this computer, but am curious what would make a faster DT computer besides destructive editing like GIMP and Photoshop.
From my work on opencl I would definitely say the most important aspect is memory size. If your images are mostly aps 4gb are fine, for 40mpix images go for 8gb
Run darktable with -d perf -d opencl.
If you see tiling → add memory. If it’s tiling on the GPU → get a GPU with more memory. If there’s no GPU → get a GPU.
It depends what you want to spend, but my NVIDIA GeForce RTX 3060 absolutely eats everything DT can throw at it, the “diffuse or sharpen” module runs very quickly even with 64 iterations and huge pixel radii.
Maybe the fact I have 32 gb RAM and an i7 processor keeps me out of trouble. I also presume in the preferences there are ways of making the previews change faster as you move sliders rather than having to wait for the change to be applied after a few moments.
I actually found NVidia was less hassle to get OpenCL working on Linux, rather than on my previous GPU, and AMD RX580 (which I sent back to Amazon, as it kept crashing and corrupting the screen).
Currently I have an Nvidia GTX1660, 6GB. No issues with darktable and opencl. Mind you, my system has an AMD Ryzen 7 3700X, and 32GB of ram.
These days, for darktable OpenCL support on Fedora Linux, it’s just a matter of installing rocm-opencl and you’re basically done. Other distros will still vary. (I’d guess Arch and Ubuntu are probably just a matter of installing a ROCm OpenCL package too?)
Installing that one package is the only extra step I had to do to get my RX 6700XT working in darktable with OpenCL acceleration on Fedora 36.
I’m glad it’s gotten smoother. I’ve heard how hard it was before. This is a very recent development. (It got so smooth just a few months ago.)
