The CUDA and OS versions are selected as follows:
- CUDA: The lastest version that has image flavour
devel
including cuDNN available. - OS: The latest version that has TensortRT libraries for
amd64
available.
ℹ️ It is taking quite a long time for these to be available forarm64
.
R images
- R: Provide NVBLAS-enabled
R_
andRscript_
.- Enabled at runtime and only if
nvidia-smi
and at least one GPU are present.
- Enabled at runtime and only if
Versions
CUDA_VERSION
Miscellaneous
CUDA_IMAGE
: The CUDA image it is derived from.CUDA_VISIBLE_DEVICES
: If unset (default), CUDA will use all available CUDA-capable devices.
👉 User-settable at runtime.
R images
- VS Code
- Extensions > R > Rterm: Linux:
/usr/local/bin/R
- Extensions > R > Rterm: Linux:
The R images use OpenBLAS by default.
To have R
and Rscript
use NVBLAS instead,
-
copy the NVBLAS-enabled executables to
~/.local/bin
for file in $(which {R,Rscript}); do cp "$file"_ "~/.local/bin/$(basename "$file")"; done
-
set Extensions > R > Rterm > Linux:
/home/USER/.local/bin/R
in VS Code settings
👉 SubstituteUSER
with your user name.
and restart the R terminal.
ℹ️ The xgboost package benefits greatly from NVBLAS, if it is installed correctly.