Start defaulting to CUDA 12 and Python 3.11 in examples

[jameslamb]

Description

RAPIDS supports CUDA 11.x and 12.x, and has for a couple of releases.

Throughout the examples in this repo, most examples default to CUDA 11. For example:

deployment/source/conf.py

Lines 27 to 42 in 26ae5ec

	versions = {
	"stable": {
	"rapids_version": stable_version,
	"rapids_container": f"nvcr.io/nvidia/rapidsai/base:{stable_version}-cuda11.8-py3.10",
	"rapids_notebooks_container": f"nvcr.io/nvidia/rapidsai/notebooks:{stable_version}-cuda11.8-py3.10",
	"rapids_conda_channels": "-c rapidsai -c conda-forge -c nvidia",
	"rapids_conda_packages": f"rapids={stable_version} python=3.10 cuda-version=11.8",
	},
	"nightly": {
	"rapids_version": f"{nightly_version}-nightly",
	"rapids_container": f"rapidsai/base:{nightly_version + 'a'}-cuda11.8-py3.10",
	"rapids_notebooks_container": f"rapidsai/notebooks:{nightly_version + 'a'}-cuda11.8-py3.10",
	"rapids_conda_channels": "-c rapidsai-nightly -c conda-forge -c nvidia",
	"rapids_conda_packages": f"rapids={nightly_version} python=3.10 cuda-version=11.8",
	},
	}

I think that in the 24.08 release, should should update the defaults in these docs to encourage the use of CUDA 12.

Benefits of this work

• nudges RAPIDS users to move to newer CUDA versions, with all the performance, usability, and other improvements that comes with

Acceptance Criteria

• all docs published from this repo use CUDA 12 in examples recommending container images and packages, except those where this is a compelling and specific reason to stay on CUDA 11

Approach

Start from here:

deployment/source/conf.py

Lines 27 to 42 in 26ae5ec

	versions = {
	"stable": {
	"rapids_version": stable_version,
	"rapids_container": f"nvcr.io/nvidia/rapidsai/base:{stable_version}-cuda11.8-py3.10",
	"rapids_notebooks_container": f"nvcr.io/nvidia/rapidsai/notebooks:{stable_version}-cuda11.8-py3.10",
	"rapids_conda_channels": "-c rapidsai -c conda-forge -c nvidia",
	"rapids_conda_packages": f"rapids={stable_version} python=3.10 cuda-version=11.8",
	},
	"nightly": {
	"rapids_version": f"{nightly_version}-nightly",
	"rapids_container": f"rapidsai/base:{nightly_version + 'a'}-cuda11.8-py3.10",
	"rapids_notebooks_container": f"rapidsai/notebooks:{nightly_version + 'a'}-cuda11.8-py3.10",
	"rapids_conda_channels": "-c rapidsai-nightly -c conda-forge -c nvidia",
	"rapids_conda_packages": f"rapids={nightly_version} python=3.10 cuda-version=11.8",
	},
	}

And then search through the code for other CUDA 11 and CUDA 12 references. Replace them and test the new references.

Notes

N/A

[raydouglass]

I'd say to also update Python to 3.11. It would be good to always promote the latest versions if possible.

[jakirkham]

This is now being done in the nightlies: #398