I can’t seem to find the page. There’s a link floating around that no longer exists, https://github.com/stan-dev/math/wiki/OpenCL-GPU-Routines. @rok_cesnovar do you know where the doc lives now? I see some references to it in the CmdStan User’s Guide but nothing that seems authoritative.
Should get this somewhere on stan-dev probably.
Thank you both! For now I’m going to edit the above wiki page to point to the new section on the Math README because it’s the top hit on Google for me.
Thanks for spotting this!
regarding GitHub wiki - the hack to find pages that are no longer there is to use a good hashtag from the past - https://github.com/stan-dev/math/wiki/OpenCL-GPU-Routines/7ec082d6c6d8524606f9de50ccb6ebe7c6664dc4 should work.
That Math wiki got edited to point to http://mc-stan.org/math/opencl_support.html which now doesn’t exist again! Anyone know where the new canonical place is?
What does GPU support speed up? is it only models involving very large matrix operations?
Yep, pretty much.
So, it wouldn’t speed up standard regression, or models like this ?
Would you mind starting another thread with your question? That will help us keep this thread on topic and make this information easier for future users to find. Thank you!
don’t know where canonical place is - it would have been part of the doxygen documentation - here’s from the tagged V 3.2.0 release (22 April 2020): https://github.com/stan-dev/math/tree/d4c32ed277a26b86bde275edce81ba1e35b508c5/doxygen/parallelism_support