[Bug 287895] graphics/nvidia*: cuda/opencl does not work on RTX 5000 series (5070) card.

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=287895

--- Comment #13 from Tomoaki AOKI <junchoon@dec.sakura.ne.jp> ---
Forgot to mention why I've installed science/linux-ai-ml-env for testing.

Searching on Internet, I found some pages state that dummy-uvm.so is wanted for
CUDA to work on FreeBSD Linuxulator, but other ports mentioned did not install
it.

And on some posts in forums and some PRs suggests to install
science/linux-ai-ml-env (or confirming the OP already installed it or not) for
CUDA regardless the target is Stable Diffusion or not. So I've installed it at
the start point for testing and found below.

% pkg which -o /compat/linux/usr/share/ai-ml-env/dummy-uvm.so
/compat/linux/usr/share/ai-ml-env/dummy-uvm.so was installed by package
science/linux-ai-ml-env
% 

Actually, Linux versions of nvidia drivers have nvidia-uvm.ko that FreeBSD
versions don't have. And some pages for Linux seems to be treating it as
prerequisite of CUDA needed to be loaded, so would be dummy-uvm.so.
But it would be just a dummy (providing symbols only, maybe?) and apps actually
require nvidia-uvm.ko to run should fail, regardless the GPU models.

-- 
You are receiving this mail because:
You are the assignee for the bug.