mirror of
https://github.com/jellyfin/jellyfin.git
synced 2026-05-04 18:09:12 +03:00
[Bug] High Ram Usage When Transcoding with NVIDIA NVENC #4767
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @LoopedSdrawkcaB on GitHub (Apr 5, 2023).
When NVIDIA NVENC is selected to transcode, memory usage slowly creeps up and uses the 16GB allocated to the VM. I have reverted back to no hardware transcoding and the memory issue disappears. When closing the stream it seems the memory usage hangs and then drops down to about 5GB of usage until I reboot the machine.
Software
Hardware
@nyanmisaka commented on GitHub (Apr 6, 2023):
@LoopedSdrawkcaB Can you share the ffmpeg transcoding log created by the stream you mentioned?
@LoopedSdrawkcaB commented on GitHub (Apr 6, 2023):
Transcode Log.txt
Here it is
@nyanmisaka commented on GitHub (Apr 6, 2023):
No errors in the log. I don't see any reason NVENC or CUDA will eat up such memory.
Can you check the RAM usage via
topornvidia-smiand find out which process is the cause?@Otako77 commented on GitHub (Apr 6, 2023):
My system is significantly weaker than yours and I have no problems.
My system:
I3-6100
8GB RAM
GTX-1650 4GB (5.25)
Ubuntu Linux, all native, no dockers
I can encode 5x UHD HDR -> Full-HD 10 Mbit or 7 other streams until the system is at its limit.
Image 1: System is idle. 1.93 RAM is used
Figure 2: System encodes 4 streams from UHD HDR->FULL-HD 10 Mbit (3 GB Ram is used)
Figure 3: System is idle again, 2GB RAM is in use
The transcoding uses about 250 MB RAM per UHD-FULL-HD stream
@nyanmisaka commented on GitHub (Apr 15, 2023):
Close as cannot reproduce on both Linux and Windows.