Hello,
I have found that on some images Darktable is slower when using OpenCL than just CPU (I have some benchmak if needed)
My CPU is Ryzen 7 5700X
I am using a Nvidia T400 2Go Vram (384 cores) that I know that is not a very fast card.
Do you think it’s normal, or is it something wrong with my config ?
If this is the 2GB model, then yes, it is expected. Basically it costs more, at some points for your GPU, to split the image into tiles that fit into the GPU’s memory and process them than it does for the CPU to process the image.
If it is the 4GB model, that should be a bit better, especially if you’re processing <24mpix images, but it can still happen with resource hungry modules.
Thank you for your reply.
I confirm it’s the 2GB model.
My RAW files are 16 Mp
I wonder now if it’s worth to enable OpenCL…
I highly depends on which modules you are using. I’ve been processing 16-Mpx-images on a 2GB GPU for a long time and it was working pretty well. Then came diffuse or sharpen which seems to require lots of memory, so thanks to tiling it was abysmally slow on the GPU (iirc, export time for images was measured in minutes). For some time I disabled OpenCL so that it was at least somewhat usable, but eventually upgraded my graphics card (now I have a second hand GTX1060 with 6 GB and no issues, even with the 24 Mpx images of my new camera).
Yes, diffuse or sharpen is pretty slow. I have set it as an automatic export style so it does not slow my process.
Thanks to a good CPU, the responsiveness of Darktable is ok for now; but if I happen to get a new camera, I will probably have to consider a new GPU.
In case you decide to upgrade the GPU and don’t want to spent too much money: Consider buying a used one, e.g. from ebay and then sell the old one. I upgraded from a GTX 760 to 1060 this way for <50 Euro (I was quite surprised that someone was still willing to pay 30 Euro for the 760…).
I keep an eye open on used GPU in my area, but it is not a priority 
Thank’s for the advice anyway.