[CI] Update NVIDIA driver to 580.82.07 (#163111)

To make CI machines capable of running CUDA-13 tests. Unfortunately, this upgrade regresses NUMBA integration, so live patch it with 6e08c9d08e

This fix was suggested in https://github.com/pytorch/pytorch/issues/162878#issuecomment-3288635745

Pull Request resolved: https://github.com/pytorch/pytorch/pull/163111
Approved by: https://github.com/huydhn

.ci/pytorch/numba-cuda-13.patch

@@ -0,0 +1,25 @@
+From 6e08c9d08e9de59c7af28b720289debbbd384764 Mon Sep 17 00:00:00 2001
+From: Michael Wang <13521008+isVoid@users.noreply.github.com>
+Date: Tue, 1 Apr 2025 17:28:05 -0700
+Subject: [PATCH] Avoid bumping certain driver API to avoid future breakage
+ (#185)
+
+Co-authored-by: isVoid <isVoid@users.noreply.github.com>
+---
+ numba_cuda/numba/cuda/cudadrv/driver.py | 3 +++
+ 1 file changed, 3 insertions(+)
+
+diff --git a/numba_cuda/numba/cuda/cudadrv/driver.py b/numba_cuda/numba/cuda/cudadrv/driver.py
+index 1641bf77..233e9ed7 100644
+--- a/numba_cuda/numba/cuda/cudadrv/driver.py
++++ b/numba_cuda/numba/cuda/cudadrv/driver.py
+@@ -365,6 +365,9 @@ def _find_api(self, fname):
+         else:
+             variants = ('_v2', '')
+ 
++        if fname in ("cuCtxGetDevice", "cuCtxSynchronize"):
++            return getattr(self.lib, fname)
++
+         for variant in variants:
+             try:
+                 return getattr(self.lib, f'{fname}{variant}')

.ci/pytorch/test.sh

@@ -32,6 +32,16 @@ if [[ "$BUILD_ENVIRONMENT" != *rocm* && "$BUILD_ENVIRONMENT" != *s390x* && -d /v
   git config --global --add safe.directory /var/lib/jenkins/workspace
 fi
 
+
+# Patch numba to avoid CUDA-13 crash, see https://github.com/pytorch/pytorch/issues/162878
+NUMBA_CUDA_DIR=$(python -c "import os;import numba.cuda; print(os.path.dirname(numba.cuda.__file__))" 2>/dev/null || true)
+if [ -n "$NUMBA_CUDA_DIR" ]; then
+  NUMBA_PATCH="$(dirname "$(realpath "${BASH_SOURCE[0]}")")/numba-cuda-13.patch"
+  pushd "$NUMBA_CUDA_DIR"
+  patch -p4 <"$NUMBA_PATCH"
+  popd
+fi
+
 echo "Environment variables:"
 env
 

.github/workflows/_linux-test.yml

@@ -169,7 +169,7 @@ jobs:
         id: install-nvidia-driver
         uses: pytorch/test-infra/.github/actions/setup-nvidia@main
         with:
-          driver-version: ${{ matrix.config == 'legacy_nvidia_driver' && '525.105.17' || '570.133.07' }}
+          driver-version: ${{ matrix.config == 'legacy_nvidia_driver' && '525.105.17' || '580.82.07' }}
         if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' && !contains(matrix.runner, 'b200') }}
 
       - name: Setup GPU_FLAG for docker run