Subject: Problems NVDEC Hardware Video Decoding Access
I'm running a video processing pipeline for sports analytics on RTX 4090 pods and need hardware-accelerated video decoding (NVDEC) for performance reasons.
TECHNICAL DETAILS:
- Pod ID: y50n3mu0w6z3nr
- GPU: RTX 4090
- Current Issue: Getting "cuvidGetDecoderCaps error 100" when trying to use NVDEC
- Library: PyNvVideoCodec 2.0.2
DIAGNOSTIC RESULTS:
- All nvdec libraries are installed correclty
- CUDA component works perfectly
- nvidia-smi shows drivers 575+
- Issue appears to be missing device access to /dev/nvidia* for video decode
VERIFICATION TEST:
I can test NVDEC access with: python -c "import PyNvVideoCodec as nvc;
decoder = nvc.SimpleDecoder('test.mp4'); decoder.get_batch_frames(1)"
This gives me this error:
Processing samples: 0it [00:00, ?it/s]
Traceback (most recent call last):
File "/app/ai-engine/matchlytics/main.py", line 1276, in <module>
main(
File "/app/ai-engine/matchlytics/main.py", line 786, in main
raise e
File "/app/ai-engine/matchlytics/main.py", line 766, in main
runner.run(
File "/app/ai-engine/matchlytics/trackers/runner.py", line 1030, in run
for sample in tqdm(frame_generator, desc="Processing samples"):
File "/usr/local/lib/python3.12/dist-packages/tqdm/std.py", line 1181, in iter
for obj in iterable:
File "/app/ai-engine/matchlytics/utils/decoders.py", line 285, in _decode_with_skipping
frame_batch = decoder.get_batch_frames(batch_size)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/PyNvVideoCodec/decoders/SimpleDecoder.py", line 115, in get_batch_frames
return self.simple_decoder.get_batch_frames(batch_size)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
_PyNvVideoCodec.PyNvVCException: HandleVideoSequence :
Error code : 100
Error Type : m_api.cuvidGetDecoderCaps(&decodecaps) returned error 100
at /project/src/VideoCodecSDKUtils/helper_classes/NvCodec/NvDecoder/NvDecoder.cpp:257
1 Reply
please someone