NVIDIA A2 Hardware Transcoding Failure - Initialization Hang on Driver 590.48 (Unraid 7.2.3)

[sebrina 0]

I'm experiencing an issue where hardware transcoding fails using my NVIDIA A2 GPUs. This happens specifically with 4K HEVC content.

System Environment:

• OS: Unraid 7.2.3
• Emby Version: 4.9.3.0 (Official Docker)
• GPU: 2x NVIDIA A2 (Tesla Ampere)
• NVIDIA Driver: 590.48 (NVIDIA Open/Standard)

Symptoms: When I start playback requiring transcoding, the client shows "No compatible streams" or "Playback Error". Checking nvidia-smi on the host shows that the ffmpeg process is created and consumes about 374MiB of VRAM, but the GPU-Util remains at 0%. The PID of ffmpeg seems to cycle/crash repeatedly.

FFmpeg Log Observation: The processing plan shows that NVDEC and NVENC are both correctly selected (WillDoInHW: True). However, there is no frame progress output (frame=...) in the logs. It seems to hang silentaly during the CUDA/OpenCL context initialization.

I have already tried:

• Changing UID/GID to 99:100.
• Giving Emby full CPU core access.
• Toggling "Experimental HEVC encoding" and "HDR Tone Mapping" (No effect).
• Intel QuickSync works fine on the same machine, so CPU performance is not the issue.

Attached are my server logs and the specific ffmpeg transcode log. Any insights would be appreciated. Thanks!

ffmpeg-transcode-8728dab4-db72-460f-9e53-7273afdf1766_1.txt embyserver.txt

[Luke 42077]

Hi, we are looking into this. Thanks.