diff --git a/.github/workflows/linux_cuda_wheel.yaml b/.github/workflows/linux_cuda_wheel.yaml
index 4e92a9095..d10c65d5f 100644
--- a/.github/workflows/linux_cuda_wheel.yaml
+++ b/.github/workflows/linux_cuda_wheel.yaml
@@ -67,10 +67,9 @@ jobs:
           # For the actual release we should add that label and change this to
           # include more python versions.
         python-version: ['3.10']
-        # We test against 12.6 to avoid having too big of a CI matrix,
+        # We test against 12.6 and 13.0 to avoid having too big of a CI matrix,
         # but for releases we should add 12.8.
-        # TODO add 13.0!
-        cuda-version: ['12.6']
+        cuda-version: ['12.6', '13.0']
         # TODO: put back ffmpeg 5 https://github.com/pytorch/torchcodec/issues/325
         ffmpeg-version-for-tests: ['4.4.2', '6', '7']
 
diff --git a/test/test_decoders.py b/test/test_decoders.py
index e68e4fe6e..ea4e91600 100644
--- a/test/test_decoders.py
+++ b/test/test_decoders.py
@@ -1224,7 +1224,7 @@ def test_full_and_studio_range_bt709_video(self, asset):
             cpu_frame = decoder_cpu.get_frame_at(frame_index).data
 
             if cuda_version_used_for_building_torch() >= (12, 9):
-                torch.testing.assert_close(gpu_frame, cpu_frame, rtol=0, atol=2)
+                torch.testing.assert_close(gpu_frame, cpu_frame, rtol=0, atol=3)
             elif cuda_version_used_for_building_torch() == (12, 8):
                 assert psnr(gpu_frame, cpu_frame) > 20
 
diff --git a/test/utils.py b/test/utils.py
index 5b6d9ea76..abe808d0f 100644
--- a/test/utils.py
+++ b/test/utils.py
@@ -73,7 +73,7 @@ def psnr(a, b, max_val=255) -> float:
 def assert_frames_equal(*args, **kwargs):
     if sys.platform == "linux":
         if args[0].device.type == "cuda":
-            atol = 2
+            atol = 3
             if get_ffmpeg_major_version() == 4:
                 assert_tensor_close_on_at_least(
                     args[0], args[1], percentage=95, atol=atol