Hi, this is just another usecase for DT. Although off-topic a bit ;-)
I have a combination of 2x16 GB and 2x8GB DDR4 RAM in my computer. Both 3200 MHz with CL16 from the same vendor but with different timings: 16-18-18-38 and 16-16-16-36. I had to experiment quite a bit to get this working. I used DT to measure RAM performance and found that it scales very good with the RAM settings. I use the following commandline to run DT on a given picture with a given xmp file: darktable-cli bench.SRW test.jpg --core --disable-opencl -d perf -d opencl The xmp file bench.SRW.xmp tells DT to execute operations which are memory expensive like: highlight reconstruction tone mapping lens correction denoise (non-local means) global tonemap equalizer local contrast The output has one line that says something like this: [dev_process_export] pixel pipeline processing took 16.472 secs (128.734 CPU) I found that the time spend in the pixel pipeline very well correlates with the RAM settings in the BIOS (if opencl is deactivated). I typically do 5+ consecutive runs and average the times. RAM Speed pixel pipeline 3200 MHz (100,0 %): 15,3 s = 100,0 % reference (only with 2x16 GB because my 2x16+2x8 did not boot at 3200 MHz) 3066 MHz ( 95,8 % = 3066 / 3200): 16,1 s = 95,0 % = 15,3 / 16,1 2933 MHz ( 91,7 %): 16,3 s = 93,9 % 2800 MHz ( 87,5 %): 17,5 s = 87,4 % 2133 MHz ( 66,7 %): 18,9 % = 80,9 % (no good correlation anymore) I found that helpful to find the best memory settings because also different CL timings can be detected with the pixel pipeline timing. Side note: I could not get it working with 3200 MHz. 3066 MHz was the first clock speed that I could boot with combined RAM and benchmark it. I though this is it and I am fine. But later I did a memtest86+ and found that this speed is throwing errors during test #7 "block move". I had to go down to 2933 MHz to get a stable setup. Kind regards Matthias ____________________________________________________________________________ darktable user mailing list to unsubscribe send a mail to [email protected]
