mpv blocks CUDA device and fails to run multiple videos using nvdec

mpv Information

mpv 0.40.0+git20250325.97cb16d68340 Copyright © 2000-2025 mpv/MPlayer/mplayer2 projects
libplacebo version: v7.349.0
FFmpeg version: 7.1 (runtime 7.1.1)
FFmpeg library versions:
   libavcodec      61.19.100 (runtime 61.19.101)
   libavdevice     61.3.100
   libavfilter     10.4.100
   libavformat     61.7.100
   libavutil       59.39.100
   libswresample   5.3.100
   libswscale      8.3.100

Other Information

- Linux version: NAME="openSUSE Tumbleweed"
- Kernel Version: Linux pc 6.14.6-1-default #1 SMP PREEMPT_DYNAMIC Tue May 13 09:56:22 UTC 2025 (ad69173) x86_64 x86_64 x86_64 GNU/Linux
- GPU Model: 01:00.0 VGA compatible controller [0300]: NVIDIA Corporation AD103 [GeForce RTX 4080] [10de:2704] (rev a1)
- Mesa/GPU Driver Version: OpenGL version string: 4.6.0 NVIDIA 570.153.02
- Window Manager and Version: plasma6-desktop-6.3.5-1.1.x86_64
- Source of mpv: official repo
- Latest known working version: N/A
- Issue started after the following happened:

Reproduction Steps

Running mpv with nvdec/cuda hwdec even as mpv --no-config --hwdec=nvdec the 1st video works fine but:

1. when I run a second video at the same time I get

[vo/gpu/cuda] cu->cuCtxCreate(&p->display_ctx, CU_CTX_SCHED_BLOCKING_SYNC, display_dev) failed -> CUDA_ERROR_DEVICE_UNAVAILABLE: CUDA-capable device(s) is/are busy or unavailable
[vo/gpu/cuda] CUDA hwdec only works with OpenGL or Vulkan backends.

2 .Also it block other CUDA software like ai model using pytorch or tensorflow.

Expected Behavior

Nvidia states that
https://docs.nvidia.com/video-technologies/video-codec-sdk/13.0/nvdec-application-note/index.html

NVDEC natively supports multiple hardware decoding contexts with negligible context-switching penalty.

So it should be possible to run in a non-blocking mode. and should be able to run multiple videos at the same time using nvdec.

Actual Behavior

1. when I run a second video at the same time I get

[vo/gpu/cuda] cu->cuCtxCreate(&p->display_ctx, CU_CTX_SCHED_BLOCKING_SYNC, display_dev) failed -> CUDA_ERROR_DEVICE_UNAVAILABLE: CUDA-capable device(s) is/are busy or unavailable
[vo/gpu/cuda] CUDA hwdec only works with OpenGL or Vulkan backends.

2 .Also it block other CUDA software like ai model using pytorch or tensorflow.

Log File

output.txt

Sample Files

No response

I carefully read all instruction and confirm that I did the following:

• I tested with the latest mpv version to validate that the issue is not already fixed.
• I provided all required information including system and mpv version.
• I produced the log file with the exact same set of files, parameters, and conditions used in "Reproduction Steps", with the addition of --log-file=output.txt.
• I produced the log file while the behaviors described in "Actual Behavior" were actively observed.
• I attached the full, untruncated log file.
• I attached the backtrace in the case of a crash.

Start with the attached output.txt and the CUDA hwdec path exercised by mpv --no-config --hwdec=nvdec. Reproduce the failure with two simultaneous videos and with another CUDA application; done means multiple videos can use nvdec without producing CUDA_ERROR_DEVICE_UNAVAILABLE or blocking other CUDA software.