特斯拉V100啥时候能用在飞牛服务器

李氏大鹏

本人有一个服务器装的飞牛系统，飞牛简明好用，业内使用非常舒服，我没有太大的需求，就需要存点自己专业的资料，最近发现最新版有飞牛龙虾，我就尝试弄一个，从网上买了性价比最高的特斯拉V100显卡（毕竟飞牛就是无头模式，这个太适合了）。在装上之后才发现，飞牛os的内核没法编译驱动，反复多次都不能解决显卡驱动的问题，很纠结，难不成就扔着了，显卡已经买了，确认收货了，好郁闷，希望官方能支持特斯拉V100显卡的支持，万分感谢

李氏大鹏

期间尝试了多次，网上找了6.12.18的内核，安装成功过，现在卡在驱动能装上，但是实时功率不能显示，显示ERR，并且在显卡调用的时候也会出现问题，好郁闷，已经折腾了最少10来天这个事情了，非常强烈需要大大们的支持

飞牛产品同学

感谢反馈，这个需求先记录下来，我们会根据评估结果推进

李氏大鹏

又没人跟我一样境遇，一起呼吁官方大佬支持哈

李氏大鹏

本人买卡为飞牛服务器增加业务能力，结果驱动打不上，求问官方是否有此显卡驱动支持计划。

我是你的小火车

我也用的V100 但是我之前已经装好新驱动了，再换的V100 能正常使用

Lostman

1. Make sure headers / build deps are in place
   On fnOS you already confirmed it’s Debian 12; now ensure the trim kernel headers exist so DKMS can build the 580 modules:

bash
sudo apt update
sudo apt install -y linux-headers-$(uname -r) build-essential dkms
On fnOS specifically that should resolve to linux-headers-6.18.18-trim; if that package isn’t found, we need to fix the Feiniu repo first, otherwise DKMS will fail to build the driver.

2. Enable the NVIDIA 580.159.04 local driver repo
   Use the .deb you linked to set up a local APT repository on the box:

bash
cd /root # or wherever you want
wget https://us.download.nvidia.com/tesla/580.159.04/nvidia-driver-local-repo-debian12-580.159.04_1.0-1_amd64.deb

sudo dpkg -i nvidia-driver-local-repo-debian12-580.159.04_1.0-1_amd64.deb
NVIDIA’s official “Debian – Driver Installation Guide” describes exactly this pattern for “Local Repository Enablement”: install the nvidia-driver-local-repo-... .deb with dpkg -i.

Then install the GPG key from the repo directory and update APT:

bash
sudo cp /var/nvidia-driver-local-repo-debian12-580.159.04/nvidia-driver-*-keyring.gpg /usr/share/keyrings/
sudo apt update
This matches what NVIDIA and Debian users are doing for 580.159.04 on Debian 12: use the local‑repo .deb, copy the keyring, then apt update.

You can sanity‑check that APT sees the repo:

bash
grep -R nvidia-driver-local-repo /etc/apt/sources.list.d
There should be a .list file created by the .deb pointing to /var/nvidia-driver-local-repo-debian12-580.159.04.

3. Install the 580 driver from that repo
   Now you choose which module flavor you want:

cuda-drivers-580 = proprietary kernel module flavor (what you want for V100)

nvidia-open-580 = open‑kernel‑module flavor (mostly for newer GPUs; V100 is better on proprietary).

So on this box I’d do:

bash

proprietary 580 branch driver (recommended for V100)

sudo apt install -y cuda-drivers-580
A Debian bug report about exactly this 580.159.04 repo shows this is the recommended command, and that it sets up DKMS for all installed kernels.

This will pull in:

nvidia-driver-580 (or similar)

the matching kernel‑dkms package

userspace libs, nvidia-smi, etc.

Then reboot:

bash
sudo reboot
After reboot, verify:

bash
nvidia-smi
modinfo nvidia | egrep 'filename|version|vermagic'
You should now see:

GPU: Tesla V100

Driver Version: 580.159.04

filename: pointing into /lib/modules/6.18.18-trim/... and a vermagic matching 6.18.18-trim.

4. CUDA toolkit on top (optional)
   That .deb/local‑repo is driver‑only; if you also want full CUDA toolkit:

Either:

add the CUDA Debian 12 repo via cuda-keyring_1.1-1_all.deb and apt install cuda-toolkit-13-0 (or similar)

Or:

install CUDA 13.x from the standalone .run as before, telling it not to install a driver.

Using NVIDIA’s CUDA repo on Debian 12 looks like:

bash
wget https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update

sudo apt install -y cuda-toolkit-13-0 # or generic 'cuda-toolkit'
This is the Debian‑centric path recommended for newer CUDA & drivers.

5. Small gotchas on fnOS
   Ensure Feiniu’s own appstore.driver.gpu.nvidia.* stuff is not conflicting; since you’ve purged all nvidia-* packages, you should be fine, but if you ever install from the fnOS App Center again it might collide with the NVIDIA 580 DKMS setup.

If you later add NVIDIA’s general CUDA repo as well, consider pinning nvidia-* packages to 580.159.04 to avoid accidentally upgrading to 590+ (which drops support for some older cards).

If you want, paste the output of:

bash
apt policy cuda-drivers-580 nvidia-open-580
nvidia-smi || echo "no nvidia-smi"
modinfo nvidia 2>/dev/null | egrep 'filename|version|vermagic' || echo "no nvidia module"
after you run through this, and I can sanity‑check that 580.159.04 is correctly bound to 6.18.18-trim on your fnOS kernel.