Can CUDA be installed on FreeBSD now ?

ZioMario · Aug 2, 2022

ok. I leave this question open for those users who are interested to apply your trick,so that potentially some other CG tools can run under FreeBSD using the CUDA libraries, like Maya or Unreal Engine 5.

Code:

# cc uvm_ioctl_override.c

ld: error: undefined symbol: main
>>> referenced by crt1_c.c:75 (/usr/src/lib/csu/amd64/crt1_c.c:75)
>>>               /usr/lib/crt1.o:(_start)
cc: error: linker command failed with exit code 1 (use -v to see invocation)

hsnyder · Aug 20, 2022

ziomario said:

Try this :

Code:

pkg install linux-nvidia-libs libc6-shim blender


copy the precompiled Cycles CUDA kernels located on the folder "/blender-3.2.1-linux-x64/3.2/scripts/addons/cycles/lib/"


of this zipped file : https://download.blender.org/release/Blender3.2/blender-3.2.1-linux-x64.tar.xz


into this folder :


/usr/local/share/blender/3.2/scripts/addons/cycles/lib


patch libc6-shim to implement openat : https://github.com/monwarez/libc6-shim/commit/7e3dffc13d33e03cd18904c6e711595b5ade72e0


git clone https://github.com/shkhln/libc6-shim.git


cd libc6-shim/bin/


env __NV_PRIME_RENDER_OFFLOAD=1 __GLX_VENDOR_LIBRARY_NAME=nvidia LD_LIBMAP="`nv-sglrun printenv LD_LIBMAP | grep -v libGL`" blender

Just for the sake of summary: I tried following the instructions in this thread and I did get it to work (or at least, my GPU shows up in blender).

Actually, I don't think you even need to build/patch libc6-shim, I'm pretty sure you can just run the `pkg install` bits from the quoted post, plus extract the precompiled cuda kernels from the blender tarball, and then use

 env __NV_PRIME_RENDER_OFFLOAD=1 __GLX_VENDOR_LIBRARY_NAME=nvidia LD_LIBMAP="`nv-sglrun printenv LD_LIBMAP | grep -v libGL`" blender

Thank you to everyone who has contributed to getting this working and documenting the process, especially shkhln and ziomario.

jbo@ · Aug 20, 2022

hsnyder If possible, it would be great if you could add additional information to your post regarding your setup such as which FreeBSD version you're running, which hardware and so on.
These things tend to change over time so someone coming along in a few years might have a better chance of figuring stuff out if these details are provided.

ZioMario · Aug 20, 2022

shkhln : is it necessary to build/patch libc6-shim ?

hsnyder · Aug 20, 2022

I'm running FreeBSD 14.0-CURRENT from about three weeks ago. Motherboard: Supermicro H11DSi, CPUs: 2x AMD Epyc 7302, GPU: 1080 Ti.

Re: patching, I was following the outlined steps above and I had already patched and built libc6-shim by the time I realized that even if you're in the libc6-shim/bin folder, simply invoking nv-sglrun instead of ./nv-sglrun will use the system installed version (from pkg) and not the one in the current directory (at least this is the case for sh and bash). Just as a sanity check, I just now deleted my patched libc6-shim and (to be extra sure) re-installed libc6-shim from pkg and it works fine. So it's definitely not required. According to freshports.org, the version of libc6-shim is the same between 13 and 14.

So the steps, then:

1. pkg install linux-nvidia-libs libc6-shim blender

2. copy the precompiled Cycles CUDA kernels located on the folder
/blender-3.2.1-linux-x64/3.2/scripts/addons/cycles/lib/
of this zipped file:
https://download.blender.org/release/Blender3.2/blender-3.2.1-linux-x64.tar.xz
into
/usr/local/share/blender/3.2/scripts/addons/cycles/lib

3. execute blender via

 env __NV_PRIME_RENDER_OFFLOAD=1 __GLX_VENDOR_LIBRARY_NAME=nvidia LD_LIBMAP="`nv-sglrun printenv LD_LIBMAP | grep -v libGL`" blender

If anybody gets nvcc working, please let me know

shkhln · Aug 20, 2022

ziomario said:
shkhln : is it necessary to build/patch libc6-shim ?

The package is up-to-date and, if you use Nvidia as your primary GPU, it should be good enough as it is. The patch from monwarez concerns hybrid dual-GPU setup (aka Optimus). nv-sglrun could also benefit from a few adjustments for Optimus, but since I'm not working on the shim at the moment, all of this can wait.

Also, currently we can't do more than a few renders in a row before running into a libc6-shim bug, requiring a Blender restart. This will be eventually™ fixed.

hsnyder said:
If anybody gets nvcc working, please let me know

Can CUDA be installed on FreeBSD now ?

Hello to everyone. I've just read this thread : https://forums.developer.nvidia.com/t/cuda-and-nv-un-register-os-un-lock-user-pages/174678/13 where has been asked to modify the code of the freebsd driver to support CUDA. I read that it has been done. So,now can we install CUDA on FreeBSD by...

forums.freebsd.org

ZioMario · Feb 16, 2023

Hello to everyone.

I've installed the nvidia driver 525.78.01 + CUDA 12 within the ubuntu linuxulator and as you can see below,both are recognized :

Code:

mario@marietto:/compat/ubuntu # sudo chroot /compat/ubuntu bash
mario@marietto:/# sudo nvidia-smi

Thu Feb 16 01:17:27 2023  
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.78.01    Driver Version: 525.78.01    CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  Off  | 00000000:01:00.0 Off |                  N/A |
| 53%   37C    P0    28W / 120W |      0MiB /  3072MiB |      2%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                         
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

now,I would also like that cycles recognizes my GPU,but it doesn't. Can someone explain to me why ? what's wrong ?

Code:

mario@marietto:/# sudo blender

Read prefs: /root/.config/blender/2.82/config/userpref.blend
ALSA lib confmisc.c:767:(parse_card) cannot find card '0'
ALSA lib conf.c:4732:(_snd_config_evaluate) function snd_func_card_driver returned error: No such fi
le or directory
ALSA lib confmisc.c:392:(snd_func_concat) error evaluating strings
ALSA lib conf.c:4732:(_snd_config_evaluate) function snd_func_concat returned error: No such file or
 directory
ALSA lib confmisc.c:1246:(snd_func_refer) error evaluating name
ALSA lib conf.c:4732:(_snd_config_evaluate) function snd_func_refer returned error: No such file or
directory
ALSA lib conf.c:5220:(snd_config_expand) Evaluate error: No such file or directory
ALSA lib pcm.c:2642:(snd_pcm_open_noupdate) Unknown PCM default
AL lib: (EE) ALCplaybackAlsa_open: Could not open playback device 'default': No such file or directo
ry
libGL error: failed to create dri screen
libGL error: failed to load driver: nouveau
/var/run/user/1001/gvfs/ non-existent directory
Warning: property 'release_confirm' not found in keymap item 'OperatorProperties'
CUDA cuInit: Unknown error
Saved session recovery to '/tmp/quit.blend'

ZioMario · Feb 16, 2023

Alain De Vos : can you figure out what could be the problem(s) ?

sko · Feb 16, 2023

why on earth would you run blender as root?

Alain De Vos · Feb 16, 2023

I'm sorry. This is not my piece of cake.
For me /compat/linux/usr is empty.

ZioMario · Feb 16, 2023

sko said:
why on earth would you run blender as root?

because it does not start as normal user :

Code:

marietto@marietto:~$ uname
Linux
marietto@marietto:~$ blender
Unable to open a display
Aborted

netchild@ · Apr 27, 2024

I'm trying to get a linux binary (biology/linux-foldingathome) to detect a nvidia card.

Without linux-nvidia-libs the linux binary (FAHClient) logs a missing cuda lib and continues without cuda. With linux-nvidia-libs, it tells "cuInit() returned 304". According to the CUDA API manual this is:

Code:

CUDA_ERROR_OPERATING_SYSTEM = 304
This indicates that an OS call failed.

The nvidia card itself is detected by the tools from linux-nvidia-libs:

Code:

# /compat/linux/usr/bin/nvidia-smi
Sat Apr 27 13:05:51 2024
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.14              Driver Version: 550.54.14      CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce GTX 1650        Off |   00000000:02:00.0 Off |                  N/A |
| 30%   37C    P0              9W /   75W |       0MiB /   4096MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+

Any ideas how to fix this?

NapoleonWils0n · Apr 27, 2024

im running the nvidia driver 550.54.14 and get exactly the same error message

when i run obs studio in a linux jail which has the matching nvidia driver installed
and try to use nvenc encoding

nvidia-smi output in the jail

Code:

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.14              Driver Version: 550.54.14      CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce GTX 1650        Off |   00000000:01:00.0 Off |                  N/A |
| N/A   41C    P8              1W /   50W |       0MiB /   4096MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+

i think there may be an issue with the 550.54.14 driver and cuda
because i did see a couple of results in google with people having the same problem

shkhln · Apr 27, 2024

netchild@ said:
Any ideas how to fix this?

I can only suggest a downgrade to 535.146.02 for now.

skywalk · May 2, 2024

The NVIDIA 535 driver is OK！
nv-sglrun nvidia-smi
shim init
Thu May 2 20:46:41 2024
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.146.02 Driver Version: 535.146.02 CUDA Version: 12.2 |
pytorch is ok：
LD_PRELOAD="/home/skywalk/work/dummy-uvm.so" python3 -c "import torch ; print(torch.randn((2,3),device='cuda'))"
tensor([[ 0.7900, -0.0157, 0.6979],
[-1.2775, -0.4350, 1.0054]], device='cuda:0')

Reference:

GitHub - verm/freebsd-stable-diffusion: Stable Diffusion on FreeBSD with CUDA support

Stable Diffusion on FreeBSD with CUDA support. Contribute to verm/freebsd-stable-diffusion development by creating an account on GitHub.

github.com

chinese doc:

安装英伟达nvidia p4计算卡驱动@FreeBSD14-CSDN博客

文章浏览阅读454次，点赞20次，收藏12次。在磕磕绊绊几天后，终于成功成功安装好nvidia p4的cuda驱动，pytorch成功运行，FreeBSD也能跑cuda AI训练拉！

skywalk.blog.csdn.net

tookie-lookie · Oct 11, 2024

Once I had a working pytorch on a FreeBSD 13.x, butt since the upgrade I hust cannot recreate the working installation. Can somebody point me in right direction ?

I've tried various nvidia driver versions, but with no luck.

Code:

(pytorch) bash-4.2# nv-sglrun nvidia-smi
shim init
Fri Oct 11 16:45:54 2024       
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.146.02             Driver Version: 535.146.02   CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  GRID M40                       Off | 00000000:04:00.0 Off |                  N/A |
|  0%   29C    P0              16W /  53W |      0MiB /  4096MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
|   1  GRID M40                       Off | 00000000:05:00.0 Off |                  N/A |
|  0%   28C    P0              16W /  53W |      0MiB /  4096MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
|   2  GRID M40                       Off | 00000000:06:00.0 Off |                  N/A |
|  0%   22C    P0              16W /  53W |      0MiB /  4096MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
|   3  GRID M40                       Off | 00000000:07:00.0 Off |                  N/A |
|  0%   23C    P0              16W /  53W |      0MiB /  4096MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
|   4  GRID M40                       Off | 00000000:86:00.0 Off |                  N/A |
|  0%   21C    P0              16W /  53W |      0MiB /  4096MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
|   5  GRID M40                       Off | 00000000:87:00.0 Off |                  N/A |
|  0%   21C    P0              16W /  53W |      0MiB /  4096MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
|   6  GRID M40                       Off | 00000000:88:00.0 Off |                  N/A |
|  0%   27C    P0              16W /  53W |      0MiB /  4096MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
|   7  GRID M40                       Off | 00000000:89:00.0 Off |                  N/A |
|  0%   24C    P0              16W /  53W |      0MiB /  4096MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
                                                                                        
+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|  No running processes found                                                           |
+---------------------------------------------------------------------------------------+
(pytorch) bash-4.2#
(pytorch) bash-4.2# python3 -c 'import torch; print(torch.cuda.is_available())'
False
(pytorch) bash-4.2# LD_PRELOAD=/home/emz/sd/dummy-uvm.so python3 -c 'import torch; print(torch.cuda.is_available())'
False

Also, the most common reason of unavailable CUDA in python is using cpu-enable pytorch instead of cuda-enabled, and I'm fully aware about this:

Code:

pytorch                   2.4.0           py3.10_cuda12.1_cudnn9.1.0_0    pytorch
ytorch-mutex             1.0                        cuda    pytorch
torchaudio                2.4.0               py310_cu121    pytorch
torchvision               0.19.0              py310_cu121    pytorch

I wonder if uvm_ioctl_override.c is still needed and whether I should use the the libc6-shim (which is more recent by 4 years) but not sure how exactly should I do it.

ZioMario · Oct 11, 2024

Try this :

GitHub - verm/freebsd-stable-diffusion: Stable Diffusion on FreeBSD with CUDA support

Stable Diffusion on FreeBSD with CUDA support. Contribute to verm/freebsd-stable-diffusion development by creating an account on GitHub.

github.com

NapoleonWils0n · Oct 11, 2024

i installed whisperx in a rocky linux jail
which has nvidia and cuda set up

however i couldnt get torch working with cuda

i have davinci resolve working in the jail so i know nvidia and cuda are working

ZioMario · Oct 11, 2024

You like to add further levels of complexity...try to make things easier.

NapoleonWils0n · Oct 11, 2024

i already had the jail set up with nvidia and cuda

you have to use the linux-miniconda-installer on freebsd to create a conda env
and have the linuxulator set up anyway

Code:

linux-miniconda-installer-0.1.0_2 Aid for installing and using miniconda

tookie-lookie · Oct 16, 2024

ZioMario said:
Try this :

"This" - what exactly ? As you can probably see, pytorch isn't seeing any CUDA devices.

freezr · Oct 16, 2024

I'd like to create a Linux Jail to run a couple of cuda application within FreeBSD, but all the threads in this forum are all spammed with a lot not-essential information:

I wonder if this guide (already posted here) is good enough:

GitHub - verm/freebsd-stable-diffusion: Stable Diffusion on FreeBSD with CUDA support

Stable Diffusion on FreeBSD with CUDA support. Contribute to verm/freebsd-stable-diffusion development by creating an account on GitHub.

github.com

ZioMario · Oct 16, 2024

Is that the same guide that I have suggested on the post #92 ?

Try this one :

https://www.reddit.com/r/freebsd/comments/1118eae/how_to_install_the_nvidia_driver_5257801_cuda_12/

freezr · Oct 16, 2024

Thanks, but it looks too much overwhelming for my tastes... ?‍?

I am going to setup a new Linux machine, it is easier... ?

akil · Dec 7, 2025

Code:

screenFetch
                                      akil@devmach
   ```                        `       OS: FreeBSD
  ` `.....---.......--.```   -/       Running Kernel: 16.0-CURRENT
  +o   .--`         /y:`      +.      Uptime: 2h 31m
   yo`:.            :o      `+-       Packages: 1136
    y/               -/`   -o/        Shell: tcsh 6.22.04
   .-                  ::/sy+:.       Resolution: 7680x2160
   /                     `--  /       DE: Enlightenment
  `:                          :`      WM: E17
  `:                          :`      WM Theme:
   /                          /       GTK Theme: Not available. [GTK2], Breeze [GTK3]
   .-                        -.       Icon Theme: oxygen
    --                      -.        Font: Not available.
     `:`                  `:`         Disk: 2.4T / 72T (3%)
       .--             `--.           CPU: AMD Ryzen 9 7950X 16-Core
          .---.....----.              GPU: NVIDIA GeForce GTX 1080
                                      RAM: 24107MiB / 65536MiB
                                     



nv-sglrun nvidia-smi
/usr/local/lib/libc6-shim/libc6.so: shim init
Sun Dec  7 18:06:09 2025      
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.105.08             Driver Version: 580.105.08     CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce GTX 1080        Off |   00000000:01:00.0  On |                  N/A |
|  0%   56C    P0             41W /  270W |     383MiB /   8192MiB |      1%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+

Cycles CUDA works on Blender 4.2, 4.5, 5.0, ffmpeg, vlc, firefox, chromium .....

Can CUDA be installed on FreeBSD now ?

​