Nvidia 5080 Testing in FreeBSD

M

mkooner

Hello,

I was able to find myself a Nvidia 5080 series and wanted to see how it worked in FreeBSD, Nvidia has claimed to launch day1 drivers for it.

They have a beta drive found here:
www.nvidia.com

Driver Details | NVIDIA

Download the <dd~LanguageName> <dd~Name> for <dd~OSName> systems. Released <dd~ReleaseDateTime>
www.nvidia.com

I was able to compile and install the driver, and it does load using the kldload nvidia-modeset command.
I am able to log into a Plasma 6 X11 session. However, it runs very slow, to open an app from click it takes 20-30seconds to open.

I am wondering if other people have had issues like this in the past with other GPU launches, and if there are any tips and tricks here I could use to speed things up.

Thanks,
MK

EDIT: If you startx with the nvidia-settings generic xconf and then close, remove the xconf file and startx again. Then X11 session works full speed with full acceleration and HDMI audio.
 
This is exciting. It's nice to have early support for the 5000 series. Hope your testing progresses well. I find the 5070 offering to be interesting pending driver tests of course. I try not invest in things unless they are tested and compatible. Will be keeping an eye on this thread for sure.
 
mkooner said:
Hello,

I was able to find myself a Nvidia 5080 series and wanted to see how it worked in FreeBSD, Nvidia has claimed to launch day1 drivers for it.

They have a beta drive found here:
www.nvidia.com

Driver Details | NVIDIA

Download the <dd~LanguageName> <dd~Name> for <dd~OSName> systems. Released <dd~ReleaseDateTime>
www.nvidia.com

I was able to compile and install the driver, and it does load using the kldload nvidia-modeset command.
I am able to log into a Plasma 6 X11 session. However, it runs very slow, to open an app from click it takes 20-30seconds to open.

I am wondering if other people have had issues like this in the past with other GPU launches, and if there are any tips and tricks here I could use to speed things up.

Thanks,
MK

EDIT: If you startx with the nvidia-settings generic xconf and then close, remove the xconf file and startx again. Then X11 session works full speed with full acceleration and HDMI audio.
Click to expand...
Does NewFeature branch of driver (565.77) work better?
And how did you install the beta (570.86.16) driver?
Via ports (x11/nvidia-driver) overriding the version? Or build directly from upstream tarball? And have you installed any of graphics/nvidia-drm-[510|515|61]-kmod or not?
 
mkooner said:
565 doesn't support Nvidia 50 series only 570 does, you need to directly download and compile it from their website with a make install
Click to expand...
I've never installed nvidia drivers directly from upstream tarball, but as ports version has mechanisms for overriding version (basically for legacy drivers), if you're lucky enough, you can use it for New Feature branch and Beta branch of drivers like 570.86.16.

For 570.86.16, x11/nvidia-driver and graphics/nvidia-drm-[510|515|61]-kmod would work with the same method described in my post in another thread. For x11/linux-nvidia-libs, some more works are needed to handle newly added files, so would take more days to file a PR for upgrading.

As upgrading ports to New Feature and/or Beta branch of driver is not allowed by the maintainer of x11/nvidia-driver, the PR would be for latest Production branch of driver, 550.144.03 with additional supports for 570.86.16.
This type of upgrades are allowed (done several times).
This is because I never file a PR for upgrading these when there's no need for additional supports to keep oportunities for additional works.

Note that you'll need 2 lines below
Code: 
hw.nvidiadrm.modeset=1
hw.nvidiadrm.fbdev=0
in your /boot/loader.conf, and adding nvidia-drm to kld_list variable in your /etc/rc.conf[.local] for 570.86.16 of Beta driver to work with X11.

If you don't set hw.nvidiadrm.fbdev=0 (or set it to 1), at least Quadro p1000 (notebook) with X11 fails to find screen (failed to acquire modesetting permission).
 
Updated info:
Found that nvidia released 570.124.04 of their driver as Production Branch of driver.

After some discussion with ashafer@ via private email, I've filed PR 285139 for updating ports and opened corresponding review D49245 as differential revision on Phablicator.

Diffs on Bugzilla and Phablicator is basically the same (the one on Phablicator unified 3 patches on Bugzilla with comments updated).
The bare-minimal version of the patch for x11/linux-nvidia-libs is part of another version as a whole.

On Linux version of the driver, VulkanSC and sandboxing supports are added among others, but not on FreeBSD native version.
But as VulkanSC seems to be targeted to autonomous vehicles and medical use-cases which most of the users don't need (would want native version if needed), I've not included it in updates for x11/linux-nvidia-libs.

For sandboxing supports, ashafer@ pointed that it relies on the containerization mechaninms on Linux, which differs from ones on FreeBSD, so I've not included it, too, as it won't work.
 
Updates for nvidia drivers for 570.124.04 landed on main branch of ports tree at commit 0de17c4dce28156a28cb89db200db0b804ebff65.
As a default in official tarball has "no screen found" problem with X11 on graphics/nvidia-drm-[510|515|61|66]-kmod, a tunable is flipped on ports side.
Anyone using Wayland only could be fine with default (I myself haven't tested, though) and setting hw.nvidiadrm.fbdev=1 in /boot/loader.conf could benefit performance.
This tunable would be flipped again (in ports side) once required functionalities are implemented in LinuxKPI and/or graphics/drm-[510|515|61|66]-kmod. See comments in diff for graphics/nvidia-drm-kmod/Makefile.common for details.
 
You must log in or register to reply here.
Back
Top