Mac OS

[robbrown]

Those are some nice compile times, maybe I should put my older smaller ATI Card back in.

Getting really tempted to ditch the Nvidia or stuff it in a Linux box with how Mojave and driver support aren’t happening.

[FarbigeWelt]

Next try: Probably It's the 1GB GPU Ram vs 24GB CPU Ram that's causing the weird results of the former test.

Splonk! Bad guessed again.

902 kSps GPU openCL and CPU together
293 kSps GPU openCL only
688 kSps CPU only

Do you see the openCL values? They should be about 2 to 3 times the listed speed according to LuxMark.

geekbench 4
HD 5750 (1.0 TFLOPS SP 32 bit, theoretical )
17.3 Gflops SFFT (fast fourier transformation, 32 MB input buffer operating in 16 KB chunks)
core i7 870 55.5 Gflops SP 32 bit (multi core), 18.4 Gflops SP (single core)
19.0 Gflops SFFT (multi core), 4.70 Gflops SFFT (single core)

Speed of this graphic card depends probably much on task complexity.

[Dade]

Rob, we added Intel Oidn as new dependency, it is available like Embree from Intel https://openimagedenoise.github.io/downloads.html for MacOS so I hope it should be too much trouble.

[robbrown]

Sounds good, just figured out my hotel charges by the day for WiFi and I’m here for 2 weeks. So may be getting creative with internet for cloning repos.

[epilectrolytics]

Splonk! Bad guessed again.

OMG how could Epilectrolytics' distant diagnostics ever get this dismally defunct??

I'll gonna spare us another try, also this doesn't belong in the developer section and you have already opened a new benchmark thread,
see ya!

[FarbigeWelt]

This doesn't belong in the developer section and you have already opened a new benchmark thread,
see ya!

Sure, I partly.agree with you, was one reason to open the benchmark thread as a generic topic.

I provided these results as feedback for the reliability of LuxCoreBlender 2.1 plugin on OS X; but also because I think there is something wrong with its openCL implementation that would need development ressources.

[epilectrolytics]

I think there is something wrong with its openCL implementation that would need development ressources.

That's very probably Apple's faulty OCL drivers which will never be fixed due to deprecation.
Everything else than a Metal transition is just wasted effort.

It's already a miracle that Rob is still providing OCL builds while everyone else, including Blender, has already given up.

[robbrown]

Ha ha, yeah the OpenCL deprecation makes me sad. Cool that it sort of works though. Lux or something else, I suspect I have Metal experience in my future.

I do now have hotel internet, sadly haven't been here much, might be able to get started with Intel Oidn this week if time permits.

[tomi-vanek]

I am trying to build the LuxCore for Mac with script ./scripts/azurepipeline/linux/build.sh in project LuxCore. Build fails - Intel Oidn is missing.

I suspect that Intel Oidn lib is missing in https://github.com/LuxCoreRender/MacOSC ... les.tar.gz, that is built from the project MacOSCompileDeps.

[Dade]

I am trying to build the LuxCore for Mac with script ./scripts/azurepipeline/linux/build.sh in project LuxCore. Build fails - Intel Oidn is missing.

I suspect that Intel Oidn lib is missing in https://github.com/LuxCoreRender/MacOSC ... les.tar.gz, that is built from the project MacOSCompileDeps.

It is very likely too miss, we are lacking a MacOS maintainer so the scripts have never been updated with the last required libraries.