CUDA.jl: Warning about loading library from system path

[CaG21]

I am using CUDA.jl on two different computers with a completely different configuration but both equipped with Windows 11. In both cases, I got a warning when loading CUDA. The warning is

┌ Warning: CUDA runtime library `nvcudart_hybrid64.dll` was loaded from a system path, `C:\WINDOWS\System32\DriverStore\FileRepository\nvtfi.inf_amd64_9bef1e779a7be7ba\nvcudart_hybrid64.dll`.
│ This may cause errors.
│
│ If you're running under a profiler, this situation is expected. Otherwise,
│ ensure that your library path environment variable (e.g., `PATH` on Windows
│ or `LD_LIBRARY_PATH` on Linux) does not include CUDA library paths.
│
│ In any other case, please file an issue.
└ @ CUDA C:\Users\cagio\.julia\packages\CUDA\Wfi8S\src\initialization.jl:218

The version of Julia is 1.11.6 while CUDA is 5.8.3. I am using an environment with only CUDA.jl installed (checked with Pkg.status()).
I checked the PATH environment variable on both systems (both user- and system-wide) and it does not contain any string related to the path C:\WINDOWS\System32\DriverStore\FileRepository. Below an example of ENV["PATH"] in one of the systems

"C:\\Program Files\\PowerShell\\7;C:\\Program Files (x86)\\Common Files\\Oracle\\Java\\java8path;C:\\Program Files (x86)\\Common Files\\Oracle\\Java\\javapath;C:\\WINDOWS\\system32;C:\\WINDOWS;C:\\WINDOWS\\System32\\Wbem;C:\\WINDOWS\\System32\\WindowsPowerShell\\v1.0\\;C:\\WINDOWS\\System32\\OpenSSH\\;C:\\Program Files\\dotnet\\;C:\\Program Files\\NVIDIA Corporation\\NVIDIA App\\NvDLISR;C:\\Program Files\\MATLAB\\R2025a\\runtime\\win64;C:\\Program Files\\MATLAB\\R2025a\\bin;C:\\Program Files\\Mosek\\11.0\\tools\\platform\\win64x86\\bin;C:\\Program Files\\PowerShell\\7\\;C:\\Program Files\\Git\\cmd;C:\\Users\\cagio\\AppData\\Local\\Microsoft\\WindowsApps;C:\\Users\\cagio\\AppData\\Local\\Programs\\Microsoft VS Code\\bin;C:\\texlive\\2025\\bin\\windows;C:\\Users\\cagio\\AppData\\Local\\Programs\\Pulsar\\resources;C:\\Users\\cagio\\AppData\\Local\\Programs\\Pulsar\\resources\\app\\ppm\\bin"

I made some tests and it seems there are no issues in using correctly CUDA.
However, I’d like to understand why I get this warning and what I can do to avoid to get that.

[eldee]

Could you post the output of CUDA.versioninfo()?

[CaG21]

Sure. Here you go

CUDA toolchain:
- runtime 13.0, artifact installation
- driver 581.15.0 for 13.0
- compiler 13.0

CUDA libraries:
- CUBLAS: 13.0.0
- CURAND: 10.4.0
- CUFFT: 12.0.0
- CUSOLVER: 12.0.3
- CUSPARSE: 12.6.2
- CUPTI: 2025.3.0 (API 130000.0.0)
- NVML: 13.0.0+581.15

Julia packages:
- CUDA: 5.8.3
- CUDA_Driver_jll: 13.0.0+0
- CUDA_Compiler_jll: 0.2.0+2
- CUDA_Runtime_jll: 0.19.0+0

Toolchain:
- Julia: 1.11.6
- LLVM: 16.0.6

1 device:
  0: NVIDIA GeForce RTX 5070 Ti Laptop GPU (sm_120, 11.544 GiB / 11.940 GiB available)

It is exactly the same for the second machine, except for the device that is

  0: NVIDIA RTX 4000 Ada Generation (sm_89, 18.779 GiB / 19.995 GiB available)

[eldee]

I’m not an expert, but that looks fine to me (though I’m a bit surprised about the 130000.0.0, I have 28.0.0 ).

Anyway, it seems like this is a known issue

System library detection should ignore `nvcudart`

with a potential fix ready

Ignore NVIDIA driver store CUDA libraries from system path warnings on Windows

So presumably the warning will go away in the next update to CUDA.jl.

[CaG21]

I completely missed the issue on github.
Thank you