UndefVarError: libcuda_original_version not defined

Hi all,
on a student’s laptop (with Nvidia P620, Windows 10) we installed CUDA.jl and faced the following error message.

julia> CUDA.versioninfo()
CUDA runtime 12.1, artifact installation
CUDA driver 12.1
NVIDIA driver 531.14.0ERROR: UndefVarError: libcuda_original_version not defined
 [1] getproperty
   @ .\Base.jl:31 [inlined]
 [2] system_driver_version
   @ C:\Users\Student\.julia\packages\CUDA\is36v\lib\cudadrv\version.jl:25 [inlined]
 [3] versioninfo(io::Base.TTY)
   @ CUDA C:\Users\Student\.julia\packages\CUDA\is36v\src\utilities.jl:46
 [4] versioninfo()
   @ CUDA C:\Users\Student\.julia\packages\CUDA\is36v\src\utilities.jl:32
 [5] top-level scope
   @ REPL[4]:1
 [6] top-level scope
   @ C:\Users\Student\.julia\packages\CUDA\is36v\src\initialization.jl:162

Is this related to “UndefVarError: libcuda not defined” as noted here:

If yes, what shall we do next? Shall we start another Julia instance in the Powershell with

$ JULIA_DEBUG=CUDA_Driver_jll julia

and run CUDA.versioninfo() again?

Could this problem also be related to a previous installation of the CUDA Toolkit?

That’s a bug, looks like I haven’t tested this edge case on Windows. I’ll have a look at a fix. In the meantime, you can always downgrade to version 3.x.

EDIT: Fix system_driver_version on platforms not supported by CUDA_Driver_jll. by maleadt · Pull Request #1854 · JuliaGPU/CUDA.jl · GitHub should fix this