This is old, but interesting.
Yes, interesting, although here either OpenCL benchmarks are (nearly) the same for both OSes or definitely won by Windows setup 
In the microphone scene, ROCm on Linux led to better performance than on Windows. The NVIDIA driver performance remained the same.
Not a dramatic difference, but a definitive win. But the important thing is that AMD drivers seemed to show different performance, while Nvidia was even.
ā¦ in 2018.
1 Like
Yep, that was 6 years ago, right now AMD should be even better than it was, especially with their commitment to open source drivers support.
And now:
1 Like
OK, no need to stay in 2018 Just went through current cross-platform OpenCL benchmarks - Luxmark 3.1 and Geekbench 6.
Geekbench 6 - Cross-Platform Benchmark
Releases Ā· LuxCoreRender/LuxMark (github.com)
And here are my results, if youād find time and could run yours on Linux - that would be interesting comparison and weād know whether differencies in DT OpenCL performance on both platforms are matter of driver or DT itself?
Do you mean I should run the tests on Linux and Nvidia?
Update:
geekbench: ASUS System Product Name - Geekbench
38061
OpenCL Score
Maybe thereās a sub-score thatās more relevant, though.
OpenCL Performance
| OpenCL Score | 38061 | |
|---|---|---|
| Background Blur | 24686 | |
| 102.2 images/sec | ||
| Face Detection | 14376 | |
| 46.9 images/sec | ||
| Horizon Detection | 48778 | |
| 1.52 Gpixels/sec | ||
| Edge Detection | 59283 | |
| 2.20 Gpixels/sec | ||
| Gaussian Blur | 56620 | |
| 2.47 Gpixels/sec | ||
| Feature Matching | 6084 | |
| 239.8 Mpixels/sec | ||
| Stereo Matching | 134798 | |
| 128.1 Gpixels/sec | ||
| Particle Physics | 92422 | |
| 4067.6 FPS |
Luxmark did not run:
kofa@eagle:/tmp/luxmark-v3.1$ ./luxmark
./luxmark.bin: error while loading shared libraries: libglut.so.3: cannot open shared object file: No such file or directory
kofa@eagle:/tmp/luxmark-v3.1$ ./luxmark.bin
./luxmark.bin: error while loading shared libraries: libembree.so.2: cannot open shared object file: No such file or directory
I do have libglut installed:
kofa@eagle:/tmp/luxmark-v3.1$ dpkg -S libglut.so.3
libglut3.12:amd64: /usr/lib/x86_64-linux-gnu/libglut.so.3.12
libglut3.12:amd64: /usr/lib/x86_64-linux-gnu/libglut.so.3.12.0
kofa@eagle:/tmp/luxmark-v3.1$ sudo apt install libglut3.12
[sudo] password for kofa:
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
libglut3.12 is already the newest version (3.4.0-1).
libglut3.12 set to manually installed.
0 upgraded, 0 newly installed, 0 to remove and 17 not upgraded.
Try going to your config folderā¦delete all the opencl kernelsā¦DT will regenerate new ones I have a feeling this has the potential to helpā¦ I have seen issues here before. Iām away and killing time on my phone but Kofa can likely tell you where they areā¦ there are in an obscure directory in your user appdata folder
1 Like
C:\Users\<user>\AppData\Local\Microsoft\Windows\INetCache\darktable (some of the path is hidden). Using a variable (for non-English Windows): %LOCALAPPDATA%\Microsoft\Windows\INetCache\darktable
1 Like
Here are my detailed results running Geekbench - still all indicate that W6600 should run faster in DT benchmarks than GTX1060:
Try going to your config folderā¦delete all the opencl kernelsā¦DT will regenerate new ones I have a feeling this has the potential to helpā¦ I have seen issues here before. Iām away and killing time on my phone but Kofa can likely tell you where they areā¦ there are in an obscure directory in your user appdata folder
Thanks, Iāll give it a shot and let know results here, but as I mentioned before I did also fresh DT install on my sonās freshly built PC with even more powerful RX 7600 and results were just slightly better, unfortunately comparable to DT+GTX1060 running on Linux.
Cleared contents of %LOCALAPPDATA%\Microsoft\Windows\INetCache\darktable directory and results are still +/- the same.
Okayā¦it was worth a shot @kofaā¦thanks for providing the location infoā¦
The only thing I tweaked that really seemed to help for my 3060ti was micronap set equal to 0 from its defaultā¦I think it might not be a suggested setting but it sped things up considerably for me and my installs over many iteration are not crashingā¦itās easy to revert also
1 Like
Thank you, Iāll try this setting too.
To be clear - Iām happy with Darktable and itās performance is OK for me on my PC I just like to benchmark new hardware when I get this 
And DT is so nice refresh after using Lightroom/Acdsee/Exposure X6 before - all these failed me - Adobe by switching to subscription-only model, Acdsee was terribly slow with Fuji RAW files, Exposure is not working properly with 4k screens and dropped development a few years ago. I was following Darktable development for years, but did not focused on this until recently, when Exposure abandoned its product. Took a time to learn Darktable basics and it is really great product now.
So thank you for your support guys, I appreciate this!
1 Like