How Does MIG Affect Gst-nvvideo4linux2 / nvv4l2decoder Usage in DeepStream

• Hardware Platform (GPU): A30
• DeepStream Version: 6.2
• Issue Type: Question

I am using an A30 sliced with MIG using config 5 (NVIDIA Multi-Instance GPU User Guide :: NVIDIA Tesla Documentation). Config 5 creates 4 slices with 1 NVDEC each.
When I run my DeepStream pipeline I am able to access instances of the nvv4l2decoder for hardware decoding on each slice.

I am curious if I were to move to an A100 which can be sliced 7 ways with config 19 (NVIDIA Multi-Instance GPU User Guide :: NVIDIA Tesla Documentation), however, each slice does not seem to have access to an NVDEC (see the table linked). Does this mean that I wouldn’t be able to use nvv4l2decoder for hardware decoding?

Physically there are 4 NVDEC cores in A30, 5 NVDEC cores in A100.

Video Encode and Decode GPU Support Matrix | NVIDIA Developer

Yes, I can see that from the specs and documents I linked.

Is my understanding correct for how nvv4l2decoders are created in DeepStream. This means that an A100 with MIG config 19 means no nvv4l2decoder elements could be created.

From software point of view, the nvv4l2decoder is working on /usr/lib/x86_64-linux-gnu/ driver. From device point of view, the nvv4l2decoder needs /dev/nvidia* nodes.

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.