Strange behaviour when installed at system level #1973
Comments
|
This is not the case for reproduction steps modified as follows: ... |
|
I am fully aware that CUDA invalidates Revise: timholy/Revise.jl#746 Code loading changed (somehow) with Julia v1.9: julia-vscode/julia-vscode#3304 |
|
Looks like julia> CUDA.versioninfo()
...
Julia packages:
- CUDA.jl: 4.3.2
- CUDA_Driver_jll: 0.5.0+1
- CUDA_Runtime_jll: 0.6.0+0
- CUDA_Runtime_Discovery: 0.2.2
(@v1.9) pkg> activate --temp
Activating new project at `/tmp/jl_VcitIx`
julia> CUDA.versioninfo()
...
Julia packages:
- CUDA.jl: 4.3.2
ERROR: KeyError: key "CUDA_Driver_jll" not found |
|
I also ran into issue #1808 and adapted my JupyterLab Julia docker stack as follows: diff --git a/base/latest.Dockerfile b/base/latest.Dockerfile
index fd7d25e..a36b3b8 100644
--- a/base/latest.Dockerfile
+++ b/base/latest.Dockerfile
@@ -9,7 +9,7 @@ ARG NB_UID=1000
ARG JUPYTERHUB_VERSION=4.0.1
ARG JUPYTERLAB_VERSION=3.6.4
ARG CODE_BUILTIN_EXTENSIONS_DIR=/opt/code-server/lib/vscode/extensions
-ARG CODE_SERVER_VERSION=4.13.0
+ARG CODE_SERVER_VERSION=4.14.0
ARG GIT_VERSION=2.41.0
ARG GIT_LFS_VERSION=3.3.0
ARG PANDOC_VERSION=3.1.1
@@ -251,7 +251,9 @@ RUN export JULIA_DEPOT_PATH=${JULIA_PATH}/local/share/julia \
&& julia -e 'using Pkg; Pkg.add(["IJulia", "Revise", "LanguageServer"]); Pkg.precompile()' \
## Install CUDA
&& if [ ! -z "$CUDA_IMAGE" ]; then \
- julia -e 'using Pkg; Pkg.add("CUDA"); Pkg.precompile()'; \
+ julia -e 'using Pkg; Pkg.add("CUDA")'; \
+ julia -e 'using CUDA; CUDA.set_runtime_version!("local")'; \
+ julia -e 'using CUDA; CUDA.precompile_runtime()'; \
fi \
&& julia -e 'using Pkg; Pkg.add(readdir("$(ENV["JULIA_DEPOT_PATH"])/packages"))' \
&& rm -rf ${JULIA_DEPOT_PATH}/registries/* \ |
|
I don't see how that's possible. If you set the preference to |
Sorry for that. Diff in wrong direction. Corrected. |
|
Yes, that's expected. See the docs, https://cuda.juliagpu.org/stable/installation/overview/#Containers. Also see this example Dockerfile: https://github.com/JuliaGPU/julia-ngc |
Describe the bug
Executing
CUDA.versioninfo()on my reference deployment results in an error:To reproduce
@maleadt This can be reproduced at https://demo.jupyter.b-data.ch:
Julia (jupyterlab/julia/pubtools:latest)CUDA.versioninfo()Manifest.toml
${JULIA_PATH}/local/share/julia/environments/v1.9/Manifest.toml:
Expected behavior
Version info
Details on Julia:
Details on CUDA:
See above.
Additional context
${JULIA_PATH}/etc/julia/startup.jl:
${HOME}/.julia/config/startup.jl:
Footnotes
If the server does not start the first time, simply give it another try. ↩
The text was updated successfully, but these errors were encountered: