Log in

View Full Version : Threadripper multi threading on Windows fixed?

ShogoXT
3rd January 2019, 10:16
https://level1techs.com/article/unlocking-2990wx-less-numa-aware-apps

https://bitsum.com/portfolio/coreprio/

Pretty much confirmed that the performance issues we're Numa issues on Windows 10 and not memory channels.

Someone who has the 32 core TR want to try this out? Wendell has been working on this for a few months at least and has videos on YouTube if you prefer that over reading. :)
huhn
3rd January 2019, 12:55
wandering threads is a know "issue" on windows.
finding the workaround is the interesting part not finding an known issue.

the real problem is if microsoft will fix/improve this anytime soon. just looking at the number of user the priority should be very low and a major windows update that works for a refresh change is more important.
nevcairiel
3rd January 2019, 13:57
NUMA is quite simply a big problem for consumer/desktop systems and software. Anyone should be fully aware of this when buying such a chip.
Atak_Snajpera
3rd January 2019, 14:53
wandering threads is a know "issue" on windows.
finding the workaround is the interesting part not finding an known issue.

the real problem is if microsoft will fix/improve this anytime soon. just looking at the number of user the priority should be very low and a major windows update that works for a refresh change is more important.

Microsoft will fix this problem immediately when intel releases own chiplet cpu. It is not a secret that those companies act like old "good" marriage ;)
huhn
4th January 2019, 03:29
then i wonder why intel CPU do perform often much worse in very high number of thread applications compared to linux where Microsoft is one of the biggest contributor. Microsoft and intel are such good friends that microsoft supports natively ARM the biggest competitor to intel so far that edge is natively written in ARM code.

if you get a workstation CPU just get a workstation OS and if this OS often linux is still incompatible with your hardware there is an easy way to fix this because these OS are often open source or atleast the kernel is.

it's the reviewer that should get some blame here.

and the next most important thing here is that noone found this "bug" before AMD. there is simply a limited for how incompetent people can treat AMD.