Hey folks, OK, this happened again, and I poked around here some more -- I needed to install Nvidia's 580 driver from the Nvidia Debian 12 repo, which compiled against the linux7 here (as some CUDA drivers stopped compiling with newer kernels -- this is a known thing: https://forums.developer.nvidia.com/t/nvidia-driver-compatibility-with-linux-kernel-6-17-debian-trixie/354354/6)
I can provide a more specific set of instructions if someone else has this issue in a timely way. Cheers! On Sat, Jul 4, 2026 at 11:41 AM Boyan Penkov <[email protected]> wrote: > > On Sat, Jul 4, 2026 at 4:25 AM didier gaumet <[email protected]> wrote: > > > > Le 03/07/2026 à 18:25, Boyan Penkov a écrit : > > [...] > > > The issue presents itself on both libgallium 25 from Trixie and > > > libgallium 26 from testing. > > > > Have you tried libgallium 25 from Trixie-backports? I have observed that > > sometimes when you use a backports kernel you need adequate backported > > libraries. > > From the modification journal of mesa-libgallium (trixie-backports): > > "[...] > > * Lower libdrm-dev to 2.4.124-1 to match the version in Trixie > > [...]" > > which suggests that mesa-libgallium from Testing is likely to create > > problems in Trixie. > > Good to know -- however, I was surprised that libdrm-dev is not > installed on my system at all, so I'll keep it in mind. > Nor is libdrm... > > > > > I would advise you against performing dist-upgrades of a stable Debian > > distro: dist-upgrades are meant only for major version upgrades, and > > even then, they are meant to take place only after an ordinary upgrade. > > Hey Didier -- your perspective here is very welcome, since I had > gathered that "dist-upgrade" > was the more general operation (which will pull in new packages as new > deps), and do it by default. > > > > > Possibly you could also assert if you need at all backports and > > fasttrack repos for other purposes, because your hardware being not the > > latest (at least the graphic card), it does not seem to need that. And > > the reliablity of a Stable Debian is greater than that of a > > Stable+backports(+fasttrack) one > > > > Again, thanks for the pointer -- I had assumed that backports was > basically as well tested as stable. > > Somehow I keep thinking that new kernel improvements (say, io_uring > marching forward), will make my CPU-based > simulations and data processing (pipeline is basically IO a terabyte > in with Python, do some bookkeeping with Python, > call a C extension to do some big array loops, do some small Python > bookkeeping, and IO ~100 MB with Python) faster, > and maybe I should back off on this... > > -- > Boyan Penkov -- Boyan Penkov
