| Summary: | Question: How will the open kernel module being the default be handled with NVIDIA 560? | ||
|---|---|---|---|
| Product: | Fedora | Reporter: | johndoe1352460 |
| Component: | nvidia-kmod | Assignee: | Nicolas Chauvet <kwizart> |
| Status: | NEW --- | ||
| Severity: | enhancement | CC: | leigh123linux, leigh123linux |
| Priority: | P1 | ||
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | GNU/Linux | ||
| namespace: | |||
|
Description
johndoe1352460
2024-05-22 18:23:45 CEST
I'm not going to switch to the open module as it doesn't support most cards. The open module will stay in tainted. Thanks for rising the point. There are two packages indeed: - nvidia-open-kmod that builds the nvidia kernel module fully from sources that only supports Turing and later GPU. This is provided as a separate component from https://github.com/NVIDIA/open-gpu-kernel-modules - nvidia-kmod extracted from the nvidia-driver that builds either: * the blob as previously seen, that supports all GPU from maxwell up to now * The pre-built nvidia-open kernel module that only works for Turing+ So for the later and for the nvidia-installer perspective, this is only an installer or build time option. But for us to correctly support Turing users it should better be a runtime option. So: 1/ Users keep install of akmod-nvidia by default 2/ Both versions are built within the same kmod-nvidia package 3/ They can be co-installed on the same system at install. 2/ Then the appropriate version is selected at runtime. (likely nvidia.ko or nvidia-open.ko and etc). I'm not sure this is realistic as it's not the way nvidia has built the driver up to now. An alternative way would be for us to use an heuristic to detect the hw and setup a script (udev rule) that will either build the nvidia blob or the pre-built nvidia-open from the nvidia-kmod package. See also https://rpmfusion.org/Howto/NVIDIA#Kernel_Open |