Transcoding makes server almost unresponsive?

[skooogis 7]

I've been running Emby on current hardware for 4-5 years, and always had the same "problem".

Let's say I have 5 Active Playback-sessions, and 3 of them are transcoding using QuickSync, the server get incredibly slow and by that I mean for the clients.
So if I start Emby on my phone, tablet or by web, first it will take like 15-20 seconds before the spinning wheel goes away, and if I browse the library It will take like 1 minute to open the library.
If I try to go to the control-panel in Emby,  either it will not load at all and just showing the "restart/shutdown button", servername and dots or it it loads extremely slow like 1 minute.
The behavior is the same from LAN and If I come from WAN.

I can not understand whats wrong or what the bottleneck is.

OSdrive is m.2
Cache-drive for transcoding etc, is a SSD.
Mediadrive is a 3.5" WD Ultrastar DC HC550
CPU is Intel Core i5 9600K
16 GB Ram
OS WS2022

From what I have googled/read people with the same CPU can have as many as 10 transcodings.
If I connect with remote desktop to check task manager, I can not really see any problem, I mean the GPU-load will jump up and down but CPU/RAM/OSdrive-load will still be almost nothing, hence why I dont understand the problem with connecting/open library/go to control-panel.

Emby is installed on OSdrive (m.2).

I have increased the DB cache-size to 1024 mb, did not make any difference.

Anyone have any ideas?

 

 

[Neminem 1518]

17 minutes ago, skooogis said:

I have increased the DB cache-size to 1024 mb, did not make any difference.

If 4.9 server set that back to 128mb 

Dev's will need logs.

[Luke 42077]

 

Hi there, let's look at an example. Please attach the information requested in how to report a media playback issue. Thanks!

 

[skooogis 7]

Yes sure, I just thought I would start a topic to see if anyone have had similar issues..
I started 4 playbacks with transcoding, and the behavior was like I've described in topic, slow connection to server with spinning wheel, slow loading library and Emby Control-panel.

embyserver.txt

[brothom 177]

8 minutes ago, skooogis said:

I started 4 playbacks with transcoding, and the behavior was like I've described in topic, slow connection to server with spinning wheel, slow loading library and Emby Control-panel.

To be fair, my server usually already has issues with a single transcoding session, let alone four.
Granted, I've upgraded my server recently but it's still not running any proper GPU so that's to be expected I suppose.

In my experience this is completely normal behaviour; you're simultaneously converting four video's from one format to another at a speed of probably around 6 MB/s (unless you have really high quality videos). That's 24 MB/s of extra overhead.

I prefer to prevent this altogether: I copy any 4k to a 1080p version (preferable h264) and I've disabled all transcoding. That way lesser devices like Chromecast can use the 1080p variant and more modern TVs and Laptop/Desktop/Whatnot can keep using the 4k versions.

[Q-Droid 989]

How busy are the drives when you're transcoding? 

The number of transcodes your CPU/iGPU can handle is very dependent on the input codec+bitrate (decoding) and the output codec+bitrate (encoding). Also the number and type of embedded subtitles can impact transcoding throughput.

Attach your transcoding logs as well to get a better idea of the actual workload.

[skooogis 7]

On 1/21/2026 at 2:21 PM, brothom said:

To be fair, my server usually already has issues with a single transcoding session, let alone four.
Granted, I've upgraded my server recently but it's still not running any proper GPU so that's to be expected I suppose.

In my experience this is completely normal behaviour; you're simultaneously converting four video's from one format to another at a speed of probably around 6 MB/s (unless you have really high quality videos). That's 24 MB/s of extra overhead.

I prefer to prevent this altogether: I copy any 4k to a 1080p version (preferable h264) and I've disabled all transcoding. That way lesser devices like Chromecast can use the 1080p variant and more modern TVs and Laptop/Desktop/Whatnot can keep using the 4k versions.

Well I have both 4K and 1080p, but clients using the Emby for Samsung need to transcode, otherwise they cant play at all sometimes..
Tried to get them to buy a Shield or Google TV Streamer etc, but as long as it works for them to play they don't really care..
However, this is more about finding out if there's a bottleneck/the cause of this behavior.

[skooogis 7]

On 1/21/2026 at 2:38 PM, Q-Droid said:

How busy are the drives when you're transcoding? 

The number of transcodes your CPU/iGPU can handle is very dependent on the input codec+bitrate (decoding) and the output codec+bitrate (encoding). Also the number and type of embedded subtitles can impact transcoding throughput.

Attach your transcoding logs as well to get a better idea of the actual workload.

The drives does not seem busy at all tbh.
The media is in general h264 @ 5000-9000 kbps
Subtitles are not embedded, they are external srt.

Attaching transcoding logs aswell.

ffmpeg-transcode-5c3c68b4-b7d4-4d32-8c4d-aa495fd74c84_1.txt ffmpeg-transcode-07c71b5e-4355-45ba-9542-9ddba4fe0c8f_1.txt ffmpeg-transcode-7a92dcea-4639-42ed-9e96-52c3e1866f73_1.txt ffmpeg-transcode-21a867c5-7757-4411-a139-5daa163538b2_1.txt ffmpeg-transcode-87b457e7-5ee2-4711-b676-d69f17926cf9_1.txt ffmpeg-transcode-a47ea198-daf7-4908-a7e7-0a05f61a5a7f_1.txt ffmpeg-transcode-c0cc7d0f-885f-42c4-8b6d-1493c6ccc5a1_1.txt ffmpeg-transcode-e4426cc3-7e36-41f7-a237-b0eff776a9eb_1.txt ffmpeg-transcode-057d694c-46a7-4a1d-b6e9-71e73df08344_1.txt ffmpeg-transcode-fe2ad180-7696-46a6-b3e5-be6e42585b08_1.txt

[horstepipe 422]

On 1/20/2026 at 9:36 PM, Neminem said:

If 4.9 server set that back to 128mb 

Dev's will need logs.

Why?

[brothom 177]

14 minutes ago, skooogis said:

Well I have both 4K and 1080p, but clients using the Emby for Samsung need to transcode, otherwise they cant play at all sometimes..
Tried to get them to buy a Shield or Google TV Streamer etc, but as long as it works for them to play they don't really care..
However, this is more about finding out if there's a bottleneck/the cause of this behavior.

Oh yes, absolutely. It's just that I tend to avoid transcoding entirely by providing a format basically any device can play (h264/aac) as a fallback.

11 minutes ago, skooogis said:

The drives does not seem busy at all tbh.
The media is in general h264 @ 5000-9000 kbps
Subtitles are not embedded, they are external srt.

That's good, having external .srt's avoids an extra drain of resources.

What about your GPU, CPU and RAM %? Can you start multiple transcoding sessions and check how much percentage of your GPU, CPU and RAM is used? If some of them are nearing 100% it would explain the bottleneck.

I can't imagine 4 transcoding sessions 36000 kbps (36 mbps) to be too much for your network, unless you're in the US or Australia ofcourse where the network connections are notoriously shoddy.

[Neminem 1518]

@horstepipe

Because things have changed and Emby is using more connection to DB.

And it was reported before 4.9 went stable.

But there is no blog about it, the default now is 20 connections each using what you have set.

If you have set 1.5 x DB size then you will run out of memory.

1GB DB size x 1.5GB x 20 = total RAM usage 30GB just for DB operations.

Really sad that there is not blog about it.

@Happy2Playcorrect me if I'm wrong.

 

Edited January 23 by Neminem

[Q-Droid 989]

Make sure your cache/transcoding SSD is being kept trimmed by Windows. Not doing so slows them down over time.

The logs don't show anything major but do have a weird pattern. For Afterburn and The Smashing Machine which were the longest playback the transcoding speed (fps rate) has a downward trend for the duration of the session. I'm curious how low it would get if you play the entire movies. I'm not saying this is "the" or "a" problem but it looks weird to me. I have no idea what changes were made to transcoding for this Emby release.

 

 

[Lessaj 467]

Maybe someone with more experience with Intel UHD graphics can chime in if this looks right but for some reason the UHD 630 is appearing twice in the codec list. Also at some random point it re-acquires the codec list but I don't see a reason for why it did that - in the past I've seen messages like the display/resolution changing, so that reprobes it, but I don't see anything here. Maybe check if the driver is crashing? Maybe it does that when you RDP? Otherwise I'm not seeing any long durations except for a handful of .ts chunks exceeding 3 seconds which may not have appeared as a problem if other chunks were delivered fast enough otherwise it would buffer during playback.

13 minutes ago, Q-Droid said:

The logs don't show anything major but do have a weird pattern. For Afterburn and The Smashing Machine which were the longest playback the transcoding speed (fps rate) has a downward trend for the duration of the session. I'm curious how low it would get if you play the entire movies. I'm not saying this is "the" or "a" problem but it looks weird to me. I have no idea what changes were made to transcoding for this Emby release.

This looks normal to me, especially with throttling enabled.

[Neminem 1518]

I also see DTS and eac3 audio transcoding, not sure how hard that is on that old CPU.

Also what kind of drive is your X drive that you use for transcoding.

Is it connected via USB.

[Lessaj 467]

13 minutes ago, Neminem said:

I also see DTS and eac3 audio transcoding, not sure how hard that is on that old CPU.

Audio transcoding is incredibly lightweight, wouldn't consider it a bottleneck at all.

[Q-Droid 989]

2 hours ago, Lessaj said:

This looks normal to me, especially with throttling enabled.

I guess it is the new normal though I hadn't noticed what the newer releases were doing. I ran a test to see if it's consistent and sure enough it is.

The playback started at full speed until it hit the first throttle event.

11:47:45.264 elapsed=00:00:00.26 frame=   37 fps=0.0 q=26.0 q=-1.0 size=N/A time=00:00:00.62 bitrate=N/A throttle=off speed=2.32x    
11:47:45.769 elapsed=00:00:00.77 frame=  167 fps=0.0 q=26.0 q=-1.0 size=N/A time=00:00:06.04 bitrate=N/A throttle=off speed=7.83x    
11:47:46.268 elapsed=00:00:01.27 frame=  291 fps=229 q=24.0 q=-1.0 size=N/A time=00:00:11.21 bitrate=N/A throttle=off speed=8.82x    
11:47:46.770 elapsed=00:00:01.77 frame=  420 fps=237 q=26.0 q=-1.0 size=N/A time=00:00:16.59 bitrate=N/A throttle=off speed=9.36x    
11:47:47.273 elapsed=00:00:02.27 frame=  550 fps=242 q=25.0 q=-1.0 size=N/A time=00:00:22.02 bitrate=N/A throttle=off speed=9.67x    
11:47:47.773 elapsed=00:00:02.77 frame=  685 fps=247 q=20.0 q=-1.0 size=N/A time=00:00:27.65 bitrate=N/A throttle=off speed=9.95x    
11:47:48.276 elapsed=00:00:03.27 frame=  822 fps=251 q=26.0 q=-1.0 size=N/A time=00:00:33.36 bitrate=N/A throttle=off speed=10.2x    
11:47:48.775 elapsed=00:00:03.77 frame=  963 fps=255 q=28.0 q=-1.0 size=N/A time=00:00:39.24 bitrate=N/A throttle=off speed=10.4x    
11:47:49.279 elapsed=00:00:04.28 frame= 1098 fps=256 q=27.0 q=-1.0 size=N/A time=00:00:44.87 bitrate=N/A throttle=off speed=10.5x    
11:47:49.779 elapsed=00:00:04.78 frame= 1224 fps=256 q=26.0 q=-1.0 size=N/A time=00:00:50.13 bitrate=N/A throttle=off speed=10.5x    
11:47:50.282 elapsed=00:00:05.28 frame= 1350 fps=255 q=25.0 q=-1.0 size=N/A time=00:00:55.38 bitrate=N/A throttle=off speed=10.5x    

After that, when throttling stopped it kept going at a slower rate.

11:48:27.426 elapsed=00:00:42.42 frame= 3774 fps= 89 q=27.0 q=-1.0 size=N/A time=00:02:36.48 bitrate=N/A throttle=off speed=3.69x    
11:48:27.925 elapsed=00:00:42.92 frame= 3899 fps= 91 q=26.0 q=-1.0 size=N/A time=00:02:41.70 bitrate=N/A throttle=off speed=3.77x    
11:48:28.425 elapsed=00:00:43.43 frame= 4030 fps= 93 q=26.0 q=-1.0 size=N/A time=00:02:47.16 bitrate=N/A throttle=off speed=3.85x    
11:48:28.926 elapsed=00:00:43.93 frame= 4164 fps= 95 q=25.0 q=-1.0 size=N/A time=00:02:52.75 bitrate=N/A throttle=off speed=3.93x    

Lower and lower with each throttle on/off cycle.

11:50:29.102 elapsed=00:02:44.05 frame= 6699 fps= 41 q=19.0 q=-1.0 size=N/A time=00:04:38.48 bitrate=N/A throttle=off speed= 1.7x    
11:50:29.550 elapsed=00:02:44.55 frame= 6807 fps= 41 q=19.0 q=-1.0 size=N/A time=00:04:42.99 bitrate=N/A throttle=off speed=1.72x    
11:50:30.050 elapsed=00:02:45.05 frame= 6935 fps= 42 q=19.0 q=-1.0 size=N/A time=00:04:48.32 bitrate=N/A throttle=off speed=1.75x    
11:50:30.551 elapsed=00:02:45.55 frame= 7065 fps= 43 q=17.0 q=-1.0 size=N/A time=00:04:53.75 bitrate=N/A throttle=off speed=1.77x    

Until it settled on a rate that was just above the video stream frame rate, to the end of playback.

12:22:39.116 elapsed=00:34:54.11 frame=52971 fps= 25 q=24.0 q=-1.0 size=N/A time=00:36:48.41 bitrate=N/A throttle=off speed=1.05x    
12:22:39.614 elapsed=00:34:54.62 frame=53096 fps= 25 q=22.0 q=-1.0 size=N/A time=00:36:53.62 bitrate=N/A throttle=off speed=1.06x    
12:22:40.117 elapsed=00:34:55.12 frame=53222 fps= 25 q=22.0 q=-1.0 size=N/A time=00:36:58.88 bitrate=N/A throttle=off speed=1.06x    
12:22:40.617 elapsed=00:34:55.62 frame=53343 fps= 25 q=24.0 q=-1.0 size=N/A time=00:37:03.93 bitrate=N/A throttle=off speed=1.06x   

 

Edited January 23 by Q-Droid

[Lessaj 467]

2 minutes ago, Q-Droid said:

I guess it is the new normal though I hadn't noticed what the newer releases were doing. I ran a test to see if it's consistent and sure enough it is.

To my knowledge it's been this way for a very long time as long as you're using throttling, I only have 30 days of logs and I updated the other day so I can't say that definitively for earlier versions but it makes sense it's always been like this. The speed calculation is basically an average rate over the whole duration so it takes a couple seconds to get up to full speed, and then if you add into the calculation a lot of time where it's not doing anything, it's going to bring that speed down closer to real time but slightly above since it's still transcoding faster than real time, the average is just brought down from the pauses.

[Q-Droid 989]

You're right. I was expecting each entry to show the work done and rate for that segment but instead it's an overall value that's averaging down as the video streams. The increment for the "frame=" value stays pretty consistent which to me indicates the actual speed is not changing.