Updating pytorch 2.10 and CUDA to 12.8 and 13.0 by swahtz · Pull Request #516 · openvdb/fvdb-core

swahtz · 2026-03-06T21:16:46Z

This pull request updates the CI and build/test/dev environments to support CUDA 13.0 and PyTorch 2.10, and streamlines related configurations. The changes ensure compatibility with newer CUDA and PyTorch versions across build, test, and publish workflows, and update dependency versions in the environment files. GCC was also updated to 14.2 to align with VFX Reference Platform 2026. Additionally, some workflow logic and build scripts are improved for flexibility and maintainability.

Build and Dependency Upgrades:

Updated cuda-version to 13.0 and pytorch-gpu to 2.10.0 in env/build_environment.yml and env/dev_environment.yml. Also upgraded gcc_linux-64 and gxx_linux-64 to 14.2, and removed or relaxed some package pins for improved compatibility. [1] [2] [3] [4]
Explicitly install torch==2.10.0 in all relevant GitHub Actions workflow steps to ensure consistent PyTorch version across CUDA 12.8 and 13.0 builds. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12]

CI Workflow Enhancements:

Added CUDA 13.0 support to the build and publish matrices, including new cuda-image-tag mappings and dynamic container image selection for both nightly and standard publish workflows. [1] [2] [3] [4]
Updated auditwheel repair steps to dynamically exclude CUDA libraries based on the CUDA major version, improving maintainability for future CUDA releases. [1] [2]
Updated Conda environment overrides and setup steps to match the new CUDA 13.0 version in documentation and workflow files. [1] [2]

Miscellaneous Improvements:

Improved the build script (build.sh) to automatically detect and use the available ninja build tool if present, enhancing build speed and reliability.
Cleaned up workflow paths and references to match updated directory structures for package builds and benchmarks. [1] [2] [3]
Added a repository check to restrict nightly publishing jobs to the main repository.

closes #511

…s/build/deploy CI Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

…ackages bucket) Updated gcc version to 14.2 to match VFX Platform 2026 Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

Update command syntax in nanovdb_editor CMake file to remove '-l' to not re-source /etc/profile and shell startup scripts which can reset PATH and makes us pickup the wrong 'ninja' Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

…a binary Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

Copilot

Pull request overview

Updates the project’s CI/build configuration and conda environments to support CUDA 13.0 and PyTorch 2.10 (plus related toolchain bumps), and improves build/publish workflow flexibility.

Changes:

Bump conda environments to cuda-version=13.0, pytorch-gpu=2.10.0, and GCC/GXX 14.2; refresh a few pins and wheel URLs.
Update GitHub Actions workflows to install torch==2.10.0 explicitly and extend publish matrices to CUDA 12.8 + 13.0.
Improve build tooling by auto-detecting Ninja and plumbing CMAKE_MAKE_PROGRAM through the build + nanovdb_editor build step.

Reviewed changes

Copilot reviewed 14 out of 14 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
`src/cmake/get_nanovdb_editor.cmake`	Passes `CMAKE_MAKE_PROGRAM` through to the nanovdb_editor wheel build.
`env/test_environment.yml`	Updates CUDA/PyTorch versions and switches TorchSparse/Scatter wheel URLs for PT 2.10 + CUDA 13.0.
`env/learn_environment.yml`	Updates CUDA/PyTorch versions and corresponding pip wheel URLs.
`env/dev_environment.yml`	Updates CUDA/PyTorch and GCC/GXX; relaxes `pybind11` pin; updates wheel URLs.
`env/build_environment.yml`	Updates CUDA/PyTorch and GCC/GXX; relaxes `pybind11` pin; trims some packages.
`build.sh`	Detects `ninja`/`ninja-build` and forwards it via `CMAKE_MAKE_PROGRAM`.
`.github/workflows/publish.yml`	Expands publish matrix to CUDA 12.8 + 13.0 and adds CUDA-image selection + dynamic auditwheel excludes.
`.github/workflows/nightly.yml`	Updates workflow paths to use repo-root `env/`, `dist/`, `tests/`, `build/` locations.
`.github/workflows/nightly-publish.yml`	Adds repo gate for nightly publish; expands matrix to CUDA 12.8 + 13.0; dynamic CUDA image selection + auditwheel excludes.
`.github/workflows/docs.yml`	Updates `CONDA_OVERRIDE_CUDA` to 13.0 and passes it into micromamba setup.
`.github/workflows/cu130.yml`	Pins `torch==2.10.0` in UV install steps for CUDA 13.0 jobs.
`.github/workflows/cu130-nightly.yml`	Pins `torch==2.10.0` in UV install steps for CUDA 13.0 nightly jobs.
`.github/workflows/cu128.yml`	Pins `torch==2.10.0` in UV install steps for CUDA 12.8 jobs.
`.github/workflows/cu128-nightly.yml`	Pins `torch==2.10.0` in UV install steps for CUDA 12.8 nightly jobs.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

Copilot · 2026-03-09T08:37:08Z

.github/workflows/cu128-nightly.yml

        run: |
          uv venv
-          uv pip install --no-cache-dir -r fvdb/env/build_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu128
+          uv pip install --no-cache-dir torch==2.10.0 -r fvdb/env/build_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu128


fvdb/env/build_requirements.txt does not exist in this repository (requirements files live under env/). This will cause the dependency install step to fail; update the path to the correct location.

Suggested change

uv pip install --no-cache-dir torch==2.10.0 -r fvdb/env/build_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu128

uv pip install --no-cache-dir torch==2.10.0 -r env/build_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu128

Copilot · 2026-03-09T08:37:09Z

.github/workflows/cu128-nightly.yml

        run: |
          uv venv
-          uv pip install --no-cache-dir -r fvdb/env/test_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu128
+          uv pip install --no-cache-dir torch==2.10.0 -r fvdb/env/test_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu128


fvdb/env/test_requirements.txt does not exist in this repository (requirements files live under env/). This will cause the dependency install step to fail; update the path to the correct location.

Suggested change

uv pip install --no-cache-dir torch==2.10.0 -r fvdb/env/test_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu128

uv pip install --no-cache-dir torch==2.10.0 -r env/test_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu128

Copilot · 2026-03-09T08:37:09Z

.github/workflows/publish.yml

+        include:
+          - cuda-version: '12.8'
+            cuda-image-tag: '12.8.1'
+          - cuda-version: '13.0'
+            cuda-image-tag: '13.0.2'


The matrix.include entries only specify cuda-version and cuda-image-tag. In GitHub Actions this adds additional matrix jobs (with missing python-version/torch-version) rather than attaching cuda-image-tag to the existing Cartesian product, which will make steps like actions/setup-python and runner labels evaluate with empty matrix values. Prefer deriving cuda-image-tag from matrix.cuda-version via an expression/step output, or switch to an include-only matrix where each included entry specifies all matrix keys (python/torch/cuda + cuda-image-tag).

Copilot · 2026-03-09T08:37:09Z

.github/workflows/nightly-publish.yml

+          - cuda-version: '12.8'
+            cuda-image-tag: '12.8.1'
+          - cuda-version: '13.0'
+            cuda-image-tag: '13.0.2'


Same matrix issue as in publish.yml: the include block here will create extra jobs that don't define python-version/torch-version, but later the workflow references those variables (runner label, setup-python, etc.). Use a mapping expression/step output for cuda-image-tag, or expand include to fully specify each matrix combination instead of mixing it with the Cartesian product.

Suggested change

- cuda-version: '12.8'

cuda-image-tag: '12.8.1'

- cuda-version: '13.0'

cuda-image-tag: '13.0.2'

- python-version: '3.10'

torch-version: '2.10'

cuda-version: '12.8'

cuda-image-tag: '12.8.1'

- python-version: '3.11'

torch-version: '2.10'

cuda-version: '12.8'

cuda-image-tag: '12.8.1'

- python-version: '3.12'

torch-version: '2.10'

cuda-version: '12.8'

cuda-image-tag: '12.8.1'

- python-version: '3.13'

torch-version: '2.10'

cuda-version: '12.8'

cuda-image-tag: '12.8.1'

- python-version: '3.10'

torch-version: '2.10'

cuda-version: '13.0'

cuda-image-tag: '13.0.2'

- python-version: '3.11'

torch-version: '2.10'

cuda-version: '13.0'

cuda-image-tag: '13.0.2'

- python-version: '3.12'

torch-version: '2.10'

cuda-version: '13.0'

cuda-image-tag: '13.0.2'

- python-version: '3.13'

torch-version: '2.10'

cuda-version: '13.0'

cuda-image-tag: '13.0.2'

Copilot · 2026-03-09T08:37:10Z

.github/workflows/cu130-nightly.yml

        run: |
          uv venv
-          uv pip install --no-cache-dir -r fvdb/env/build_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu130
+          uv pip install --no-cache-dir torch==2.10.0 -r fvdb/env/build_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu130


fvdb/env/build_requirements.txt does not exist in this repository (the requirements files live under env/). This will cause the job to fail when installing dependencies. Update the path to the correct location (and keep it consistent with the other workflows updated in this PR).

Suggested change

uv pip install --no-cache-dir torch==2.10.0 -r fvdb/env/build_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu130

uv pip install --no-cache-dir torch==2.10.0 -r env/build_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu130

Copilot · 2026-03-09T08:37:10Z

.github/workflows/cu130-nightly.yml

        run: |
          uv venv
-          uv pip install --no-cache-dir -r fvdb/env/test_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu130
+          uv pip install --no-cache-dir torch==2.10.0 -r fvdb/env/test_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu130


fvdb/env/test_requirements.txt does not exist in this repository (the requirements files live under env/). This will break dependency installation in the nightly test job; update the path to the correct location.

Suggested change

uv pip install --no-cache-dir torch==2.10.0 -r fvdb/env/test_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu130

uv pip install --no-cache-dir torch==2.10.0 -r env/test_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu130

First pass updating pytorch 2.10 and CUDA to 12.8 and 13.0 across env…

6f797fc

…s/build/deploy CI Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

swahtz added this to the v0.4 milestone Mar 6, 2026

swahtz self-assigned this Mar 6, 2026

swahtz added CI Issues related to the Github actions CI/CD. For build issues use CMake/Build cmake/build Issues related to building and CMake labels Mar 6, 2026

Change torchsparse/torchscatter packages to updated builds (at fvdb-p…

7ebcd96

…ackages bucket) Updated gcc version to 14.2 to match VFX Platform 2026 Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

swahtz marked this pull request as ready for review March 9, 2026 06:21

swahtz requested a review from a team as a code owner March 9, 2026 06:21

swahtz requested review from fwilliams and phapalova March 9, 2026 06:21

swahtz added 7 commits March 9, 2026 19:22

fix nightly-publish to only run in openvdb repo

8d4416c

Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

Resolving CONDA_OVERRIDE_CUDA env ssetting

32429c1

Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

Merge branch 'main' into issue-511-pytorch-2-10

d92eb5a

Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

remove pybind11 pin and remove pybind11 from runtime environments

68027b1

Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

remove unneeded python libs

f5db087

Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

fix issue with get_nanovdb_editor pip install finding the system ninj…

3014db1

…a binary Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>

swahtz requested a review from Copilot March 9, 2026 08:31

Copilot started reviewing on behalf of swahtz March 9, 2026 08:32 View session

Copilot AI reviewed Mar 9, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Updating pytorch 2.10 and CUDA to 12.8 and 13.0#516

Updating pytorch 2.10 and CUDA to 12.8 and 13.0#516
swahtz wants to merge 9 commits intoopenvdb:mainfrom
swahtz:issue-511-pytorch-2-10

swahtz commented Mar 6, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Mar 9, 2026

Uh oh!

Copilot AI Mar 9, 2026

Uh oh!

Copilot AI Mar 9, 2026

Uh oh!

Copilot AI Mar 9, 2026

Uh oh!

Copilot AI Mar 9, 2026

Uh oh!

Copilot AI Mar 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	uv pip install --no-cache-dir torch==2.10.0 -r fvdb/env/build_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu128
	uv pip install --no-cache-dir torch==2.10.0 -r env/build_requirements.txt --extra-index-url https://download.pytorch.org/whl/cu128

Conversation

swahtz commented Mar 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 9, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

swahtz commented Mar 6, 2026 •

edited

Loading