CUDA not working

[Flock1]

Hi,

I am very new to Julia, hence please don't mind the naiveness of the query. I have installed Julia and installed the CUDA package. However, when I try to test or check the version, I get an error:

 CUDA.version()
┌ Error: Could not initialize CUDA
│   exception =
│    could not load library "libcuda"
│    libcuda.so: cannot open shared object file: No such file or directory
│    Stacktrace:
│     [1] dlopen(::String, ::UInt32; throw_error::Bool) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.5/Libdl/src/Libdl.jl:109
│     [2] dlopen at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.5/Libdl/src/Libdl.jl:109 [inlined]
│     [3] #dlopen#2 at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.5/Libdl/src/Libdl.jl:105 [inlined]
│     [4] dlopen at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.5/Libdl/src/Libdl.jl:105 [inlined] (repeats 2 times)
│     [5] (::CUDA.var"#2956#cache_fptr!#617")() at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/utils/call.jl:31
│     [6] macro expansion at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/utils/call.jl:39 [inlined]
│     [7] __configure__(::Bool) at /home/sarvagya/.julia/packages/CUDA/dZvbp/src/initialization.jl:90
│     [8] (::CUDA.var"#609#610"{Bool})() at /home/sarvagya/.julia/packages/CUDA/dZvbp/src/initialization.jl:29
│     [9] lock(::CUDA.var"#609#610"{Bool}, ::ReentrantLock) at ./lock.jl:161
│     [10] _functional(::Bool) at /home/sarvagya/.julia/packages/CUDA/dZvbp/src/initialization.jl:26
│     [11] functional(::Bool) at /home/sarvagya/.julia/packages/CUDA/dZvbp/src/initialization.jl:19
│     [12] libcuda at /home/sarvagya/.julia/packages/CUDA/dZvbp/src/initialization.jl:50 [inlined]
│     [13] (::CUDA.var"#676#cache_fptr!#7")() at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/utils/call.jl:31
│     [14] macro expansion at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/utils/call.jl:39 [inlined]
│     [15] macro expansion at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/cudadrv/libcuda.jl:28 [inlined]
│     [16] macro expansion at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/cudadrv/error.jl:108 [inlined]
│     [17] cuDriverGetVersion(::Base.RefValue{Int32}) at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/utils/call.jl:93
│     [18] version() at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/cudadrv/version.jl:10
│     [19] top-level scope at REPL[2]:1
│     [20] eval(::Module, ::Any) at ./boot.jl:331
│     [21] eval_user_input(::Any, ::REPL.REPLBackend) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.5/REPL/src/REPL.jl:134
│     [22] repl_backend_loop(::REPL.REPLBackend) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.5/REPL/src/REPL.jl:195
│     [23] start_repl_backend(::REPL.REPLBackend, ::Any) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.5/REPL/src/REPL.jl:180
│     [24] run_repl(::REPL.AbstractREPL, ::Any; backend_on_current_task::Bool) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.5/REPL/src/REPL.jl:292
│     [25] run_repl(::REPL.AbstractREPL, ::Any) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.5/REPL/src/REPL.jl:288
│     [26] (::Base.var"#806#808"{Bool,Bool,Bool,Bool})(::Module) at ./client.jl:399
│     [27] #invokelatest#1 at ./essentials.jl:710 [inlined]
│     [28] invokelatest at ./essentials.jl:709 [inlined]
│     [29] run_main_repl(::Bool, ::Bool, ::Bool, ::Bool, ::Bool) at ./client.jl:383
│     [30] exec_options(::Base.JLOptions) at ./client.jl:313
│     [31] _start() at ./client.jl:506
└ @ CUDA ~/.julia/packages/CUDA/dZvbp/src/initialization.jl:97
ERROR: AssertionError: CUDA.jl did not successfully initialize, and is not usable.
Stacktrace:
 [1] libcuda at /home/sarvagya/.julia/packages/CUDA/dZvbp/src/initialization.jl:50 [inlined]
 [2] (::CUDA.var"#676#cache_fptr!#7")() at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/utils/call.jl:31
 [3] macro expansion at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/utils/call.jl:39 [inlined]
 [4] macro expansion at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/cudadrv/libcuda.jl:28 [inlined]
 [5] macro expansion at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/cudadrv/error.jl:108 [inlined]
 [6] cuDriverGetVersion(::Base.RefValue{Int32}) at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/utils/call.jl:93
 [7] version() at /home/sarvagya/.julia/packages/CUDA/dZvbp/lib/cudadrv/version.jl:10
 [8] top-level scope at REPL[2]:1

What should I do? NVIDIA driver and CUDA are installed on my machine.

[maleadt]

libcuda.so: cannot open shared object file: No such file or directory

It doesn't find libcuda, are you sure the NVIDIA driver is properly installed?
And you don't need to install CUDA, CUDA.jl will provide its own.

[Flock1]

I think it's installed properly because this is what I get when I run nvidia-smi:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.100      Driver Version: 440.100      CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+

[maleadt]

nvidia-smi working is unrelated to libcuda availability. Did you look in your system for libcuda? It needs to be installed in a location that can be found by the dynamic linker, if not your system is badly configured (i.e. Libdl.find_library("libcuda") should return something). You can force a directory to be searched in by setting LD_LIBRARY_PATH, but generally you need an entry in /etc/ld.so.conf.d if libcuda isn't installed in a default location.

[Flock1]

I search for libcuda and these are some of the results I got:

/usr/lib/i386-linux-gnu/libcuda.so
/usr/lib/i386-linux-gnu/libcuda.so.1
/usr/lib/i386-linux-gnu/libcuda.so.440.100
/usr/lib/x86_64-linux-gnu/libcuda.so.1
/usr/lib/x86_64-linux-gnu/libcuda.so.440.100

[maleadt]

Please try my other suggestions...

[Flock1]

For Libdl.find_library("libcuda"), I got this output:
bash: syntax error near unexpected token '"libcuda"'

[maleadt]

That's a Julia command. https://docs.julialang.org/en/v1/stdlib/Libdl/#Libdl.find_library
Open a REPL and start with using Libdl

[Flock1]

Hey, apologies. I am so new to Julia that keep using import instead of using.

This is what I got when I ran the command: "".

My LD_LIBRARY_PATH is set to :/usr/lib/i386-linux-gnu/libcuda.so

[maleadt]

My LD_LIBRARY_PATH is set to :/usr/lib/i386-linux-gnu/libcuda.so

That's wrong, you should set it do the directory. Furthermore, CUDA is only supported on 64-bit, so you need to use the other directory.

And finally, the LD_LIBRARY_PATH is a temporary solution. /usr/lib/x86_64-linux-gnu should have been added to the library search path persistently with an entry in /etc/ld.so.conf.d

Since neither of these are Julia or CUDA.jl issues, I'll close this issue. If you have more problems with your system, I suggest you hop by Slack (#gpu or #helpdesk channel), or make a post on Discourse 🙂