Please read this before reporting a bug:
https://wiki.archlinux.org/title/Bug_reporting_guidelines
Do NOT report bugs when a package is just outdated, or it is in the AUR. Use the 'flag out of date' link on the package page, or the Mailing List.
REPEAT: Do NOT report bugs for outdated packages!
https://wiki.archlinux.org/title/Bug_reporting_guidelines
Do NOT report bugs when a package is just outdated, or it is in the AUR. Use the 'flag out of date' link on the package page, or the Mailing List.
REPEAT: Do NOT report bugs for outdated packages!
FS#69007 - [python-tensorflow-cuda] re-add compute capability 3.5
Attached to Project:
Community Packages
Opened by Martin Saraceno (tinux) - Wednesday, 16 December 2020, 21:51 GMT
Last edited by Sven-Hendrik Haase (Svenstaro) - Saturday, 19 December 2020, 11:58 GMT
Opened by Martin Saraceno (tinux) - Wednesday, 16 December 2020, 21:51 GMT
Last edited by Sven-Hendrik Haase (Svenstaro) - Saturday, 19 December 2020, 11:58 GMT
|
DetailsI would highly appreciate if compute capability 3.5 were added to the package again. I am using a Titan Black and cannot use TF without it.
Thank you in advance for the consideration. |
This task depends upon
Closed by Sven-Hendrik Haase (Svenstaro)
Saturday, 19 December 2020, 11:58 GMT
Reason for closing: Won't fix
Saturday, 19 December 2020, 11:58 GMT
Reason for closing: Won't fix
FS#69005? If not, what specifically was disabled, and when?FS#69005). The problem that I am facing is that TF is compiled with `TF_CUDA_COMPUTE_CAPABILITIES=5.2,5.3,6.0,6.1,6.2,7.0,7.2,7.5,8.0,8.6`. If I change this to `TF_CUDA_COMPUTE_CAPABILITIES=3.5` in PKGBUILD and recompile, only then there will be a kernel available for my hardware (Nvidia Titan Black). I tried this and it works. If not **nothing** will run on the GPU.I assume that changing the PKGBUILD to `export TF_CUDA_COMPUTE_CAPABILITIES=3.5,5.2,5.3,6.0,6.1,6.2,7.0,7.2,7.5,8.0,8.6` should solve this generally. I am not sure what other compute capabilities, other than 3.5 which I would greatly appreciate, would be useful for other people. I don't know if there are any drawback associated to this, but I guess none.