![]() |
![]() |
#1 |
Aug 2002
21C316 Posts |
![]() |
![]() |
![]() |
![]() |
#2 | |
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
2×17×227 Posts |
![]() Quote:
CUDA, OpenGL, DirectML, but no OpenCL. https://github.com/Microsoft/WSL/issues/3789 Maybe someday. |
|
![]() |
![]() |
![]() |
#3 |
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
2×17×227 Posts |
![]()
Only mention I've seen of ROCm on Windows10 as a possibility:
https://github.com/ROCm-Developer-Tools/HIP/issues/84 toward the end; Github |
![]() |
![]() |
![]() |
#4 | |
"Teal Dulcet"
Jun 2018
6C16 Posts |
![]() Quote:
|
|
![]() |
![]() |
![]() |
#5 |
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
2×17×227 Posts |
![]()
Thanks for the links tdulcet.
Sounds like mapping or emulation, inefficient, slow, to me. Not OpenCL native support, but a translation layer into DirectX. https://devblogs.microsoft.com/direc...rs-to-directx/ Perhaps better than nothing, wait and see, until whenever; no schedule given in the provided links, for the availability of the mapping layer, or for the required WSL support to follow. I'm skeptical that MS will map a sufficient, compatible, and efficient subset of OpenCL to preserve gpuowl and mfakto performance close to that on native OpenCL drivers, at least in the initial years. MS generally operates by successive approximation over several major releases. My impression over nearly 40 years of use. "initial support targeting OpenCL 1.2" probably means no 2.0 atomics used by Gpuowl since after ~v1.9. https://devblogs.microsoft.com/direc...rs-to-directx/ https://mersenneforum.org/showpost.p...&postcount=691 https://mersenneforum.org/showpost.p...postcount=2025 Perhaps Preda would consider someday splitting to two back ends again, this time OpenCL and DirectX https://mersenneforum.org/showpost.p...&postcount=429 Last fiddled with by kriesel on 2021-04-25 at 13:44 |
![]() |
![]() |