I've been wondering if we want to switch the default value of DRM_NOUVEAU_GSP_DEFAULT to y for the nouveau kernel module. My impression is that at this point the GSP code path is both much better tested and much faster than the classic firmware code paths on turing and ampere. I think that encouraging distributions to use GSP by default will go a long way to improving the default experience for users. Thoughts? Is there any reason not to do this? -------------- next part -------------- An HTML attachment was scrubbed... URL: <https://lists.freedesktop.org/archives/nouveau/attachments/20250616/cb2330bd/attachment.htm>
On 6/16/25 21:41, M Henning wrote:> I've been wondering if we want to switch the default value > of?DRM_NOUVEAU_GSP_DEFAULT to y for the nouveau kernel module. My > impression is that at this point the GSP code path is both much better > tested and much faster than the classic firmware code paths on turing > and ampere. I think that encouraging distributions to use GSP by default > will go a long way to improving the default experience for users. > > Thoughts? Is there any reason not to do this?Yes please :) When the GSP did not bring anything to users because the userspace was unable to really make use of it, I think it made sense to keep it disabled. The situation has however changed quite drastically and at this point, not using the GSP has a big impact on usability. And users who do not wish to depend on a proprietary firmware can always use `nouveau.config=NvGspRm=0` to disable it (if they can't find a Kepler GPU that would be supported using only open source firmware). I would however appreciate better error reporting when loading the GSP fails, as it can currently be quite unhelpful without increasing the drm log verbosity: [ 2.177205] nouveau 0000:01:00.0: gsp ctor failed: -2 Cheers, Martin
On 6/17/25 04:41, M Henning wrote:> I've been wondering if we want to switch the default value > of?DRM_NOUVEAU_GSP_DEFAULT to y for the nouveau kernel module. My > impression is that at this point the GSP code path is both much better > tested and much faster than the classic firmware code paths on turing > and ampere. I think that encouraging distributions to use GSP by > default will go a long way to improving the default experience for users. > > Thoughts? Is there any reason not to do this?I have no objection to this.? Initially, the primary reason for keeping it disabled was to give some time to shake out any regressions - but I think we're well past that point now. Ben.