CUDA toolkit layer for OpenCharly GPU images — Fedora and Arch.
The cuda candy installs the CUDA compiler driver (nvcc), the CUDA runtime/dev
libraries (cudart, nvrtc, cupti, curand, cccl, cufile), cuDNN, and
onnxruntime, stitched into the canonical /usr layout on both Fedora and Arch.
It depends on the nvidia candy for GPU runtime support and on ffmpeg for
codec libraries.
On Fedora the RPMs land in /usr directly. On Arch the packages install under
/opt/cuda, so post-install tasks symlink the binaries, headers, and libraries
into the same /usr-rooted paths. With that stitch in place, every downstream
GPU consumer compiles against one layout regardless of distro. Each artifact is a
real file or library on disk, so its presence is directly verifiable.
| Property | Value |
|---|---|
| Layer / candy | cuda |
| Requires | layer-nvidia, layer-ffmpeg |
| Environment | CUDA_HOME=/usr |
| Binaries | /usr/bin/nvcc (+ cuobjdump, ptxas, ...) |
| Headers | /usr/include/cudnn.h, CUDA headers |
| Libraries | /usr/lib64/libcudart.so*, libcudnn, libcurand, ... |
| Service / port | none |
Per-distro packages:
fedora—cuda-nvcc,cuda-cudart-devel,cuda-cudart-static,cuda-nvrtc-devel,cuda-cupti-devel,cuda-cccl-devel,cuda-cudnn,libcurand-devel,libcufile-devel,onnxruntime,libaio-devel,cpio(NVIDIA's CUDA repo + negativo17fedora-multimediaforonnxruntime).arch—cuda,cudnn,python-onnxruntime-cpu.
Compose the layer by pinning this repo in a box's candy: list:
my-gpu-box:
candy:
base: fedora-nonfree
candy:
- '@github.com/opencharly/layer-cuda:v2026.243.0408'The nvidia dependency supplies the runtime; cuda adds the toolkit. After the
image is built (no physical GPU needed to verify the artifacts):
nvcc --version
ls /usr/include/cudnn.h
ls /usr/lib64/libcudart.so*charly.yml— thecuda:candy entity: therequire:deps, the per-distro package arms and repos, theCUDA_HOMEenv, the stitch + extractplan:steps, and thecheck:assertions, plus the embeddedskill:entity.CHANGELOG/— per-CalVer release notes..github/workflows/tag-on-merge.yml— CalVer tag +CHANGELOG/on merge.README.md— this user overview.
- Owning skill:
/charly-distros:cuda— the CUDA toolkit, cuDNN, and ONNX Runtime reference - GPU runtime:
/charly-distros:nvidia - Codecs:
/charly-selkies:ffmpeg - Derived GPU boxes:
/charly-languages:python-ml,/charly-jupyter:jupyter,/charly-ollama:ollama,/charly-comfyui:comfyui,/charly-immich:immich-ml opencharly/charly— the charly CLI and image builderopencharly/opencharly— the umbrella