Skip to content

Fix documentation for setting TORCH_CUDA_ARCH_LIST #328

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
matthewdcong merged 2 commits into from
Nov 12, 2025
Merged

Fix documentation for setting TORCH_CUDA_ARCH_LIST #328

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9851966
Fix documentation for setting TORCH_CUDA_ARCH_LIST
matthewdcong Nov 12, 2025
952659d
Add support for TORCH_CUDA_ARCH_LIST
matthewdcong Nov 12, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 2 additions & 3 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,8 +106,7 @@ docker build -t fvdb-devel .
docker run -it --mount type=bind,src="$(pwd)",target=/workspace fvdb-devel bash
cd /workspace;
pip install -r env/build_requirements.txt
TORCH_CUDA_ARCH_LIST="7.5;8.0;9.0;10.0;12.0+PTX" \
./build.sh install -v
./build.sh install --cuda-arch-list="7.5;8.0;9.0;10.0;12.0+PTX" -v
```

In order to extract an artifact from the container such as the Python wheel, query the container ID using `docker ps` and copy the artifact using `docker cp`.
Expand All @@ -126,7 +125,7 @@ Using a Python virtual environment enables you to use your system provided compi
python -m venv fvdb
source fvdb/bin/activate
pip install -r env/build_requirements.txt
TORCH_CUDA_ARCH_LIST="7.5;8.0;9.0;10.0;12.0+PTX" ./build.sh install -v
./build.sh install --cuda-arch-list="7.5;8.0;9.0;10.0;12.0+PTX" -v
```

Note: adjust the TORCH_CUDA_ARCH_LIST to suit your needs. If you are building just to run on a single machine, including only the present GPU architecture(s) reduces build time.
Expand Down
11 changes: 5 additions & 6 deletions build.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@ usage() {
echo " -h, --help Display this help message and exit."
echo " --cuda-arch-list <value> Set TORCH_CUDA_ARCH_LIST (auto-detects if not specified; "
echo " use 'default' to force auto-detect)."
echo " Example: --cuda-arch-list=8.0;8.6+PTX"
echo " Example: --cuda-arch-list=\"8.0;8.6+PTX\""
echo ""
echo "Build Modifiers (for 'install' and 'wheel' build types, typically passed after build_type):"
echo " gtests Enable building tests (sets FVDB_BUILD_TESTS=ON)."
Expand Down Expand Up @@ -85,18 +85,17 @@ setup_parallel_build_jobs() {
set_cuda_arch_list() {
local list="$1"
if [ -n "$list" ] && [ "$list" != "default" ]; then
echo "Using TORCH_CUDA_ARCH_LIST=$list"
echo "Using specified TORCH_CUDA_ARCH_LIST=$list"
export TORCH_CUDA_ARCH_LIST="$list"
else
if ([ "$list" == "default" ] || [ -z "$TORCH_CUDA_ARCH_LIST" ]) && command -v nvidia-smi >/dev/null 2>&1; then
if ([ "$list" == "default" ] && [ -z "$TORCH_CUDA_ARCH_LIST" ]) && command -v nvidia-smi >/dev/null 2>&1; then
echo "Detecting CUDA architectures via nvidia-smi"
# Try via nvidia-smi (compute_cap available on newer drivers)
TORCH_CUDA_ARCH_LIST=$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader 2>/dev/null | awk 'NF' | awk '!seen[$0]++' | sed 's/$/+PTX/' | paste -sd';' -)
fi

if [ -n "$TORCH_CUDA_ARCH_LIST" ]; then
export TORCH_CUDA_ARCH_LIST
echo "Detected CUDA architectures: $TORCH_CUDA_ARCH_LIST"
elif ([ "$list" == "default" ] && [ -n "$TORCH_CUDA_ARCH_LIST" ]); then
echo "Using environment TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST"
else
echo "Warning: Could not auto-detect CUDA architectures. Consider setting TORCH_CUDA_ARCH_LIST manually (e.g., 8.0;8.6+PTX)."
fi
Expand Down
Loading