Advice Needed on Adding GPU for Emby Transcoding

[Dalpuri 18]

Hi Emby Community,

I am currently running an Emby server on a system with the following specifications:

• Processor: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
• RAM: 128 GB
• Machine Type: Rack Mount Chassis
• Operating System: Ubuntu 20.04.6 LTS
• Storage: HP P440, HP P840ar, HP LOGICAL VOLUME (multiple disks)
• Display Resolution: 1024x768 pixels

My Emby server primarily handles transcoding for both LAN and some remote network devices. However, I've been facing significant performance issues during transcoding, where CPU usage spikes to 90% or higher, making the server laggy. On average, there are 20-30 active devices, but this number can sometimes exceed 100.

Given this situation, I'm considering adding a dedicated GPU to improve transcoding efficiency. I've heard that Nvidia GPUs are highly recommended for this purpose due to their NVENC support. Here are my specific questions:

• Will adding a dedicated GPU help with my transcoding performance?
• Which Nvidia GPU model would you recommend for handling a high number of simultaneous streams, including some 4K streams?
• Are there any particular configuration steps or considerations I should be aware of when setting up the GPU for Emby transcoding?

I've seen recommendations for various Nvidia GPUs such as the Quadro P2000, GeForce GTX 1660 Super, and RTX series. My goal is to find a balance between performance and cost, ensuring that the server can handle multiple 1080p and 4K streams efficiently.

I appreciate any insights, recommendations, or experiences you can share regarding the best GPU choice and setup for this scenario.

Thank you!

[Neminem 622]

Wow that's a lot of users for your personal media emby server.
Bringing all of your home videos, music, and photos together into one place has never been easier. Your personal Emby Server automatically converts and streams your media on-the-fly to play on any device.

51 minutes ago, Dalpuri said:

On average, there are 20-30 active devices, but this number can sometimes exceed 100.

[Dalpuri 18]

3 hours ago, jaycedk said:

Wow that's a lot of users for your personal media emby server.
Bringing all of your home videos, music, and photos together into one place has never been easier. Your personal Emby Server automatically converts and streams your media on-the-fly to play on any device.

Well, it's a media center for some people in a fixed area under the same network hehe. Using it this way for the convenience. 

[Lessaj 198]

Depending on how many concurrent transcode streams there are it may be ideal to look into a P2000 (or P2200) because it has unlimited streams however it's quite dated at this point and a more modern Turing or Ampere based model would yield better performance - consumer grade cards are limited to 8 stream but there is a driver workaround available to get unlimited streams for consumer grade cards. It's been some time since I had a P2000 in my system, it's still a very capable transcoding card for multiple streams but I'm not sure it would be suitable for more than 2 4K streams at the same time, especially if tone mapping. Having said that really just buy the best card you can afford that fits and you have power budget for, especially with throtttling enabled it shouldn't be much of an issue regardless of what card you get.

https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new

Edited June 22 by Lessaj

[Dalpuri 18]

15 hours ago, Lessaj said:

Depending on how many concurrent transcode streams there are it may be ideal to look into a P2000 (or P2200) because it has unlimited streams however it's quite dated at this point and a more modern Turing or Ampere based model would yield better performance - consumer grade cards are limited to 8 stream but there is a driver workaround available to get unlimited streams for consumer grade cards. It's been some time since I had a P2000 in my system, it's still a very capable transcoding card for multiple streams but I'm not sure it would be suitable for more than 2 4K streams at the same time, especially if tone mapping. Having said that really just buy the best card you can afford that fits and you have power budget for, especially with throtttling enabled it shouldn't be much of an issue regardless of what card you get.

https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new

If i want the video to be transcoded almost instantly assuming I'll only have 1080p max videos. Should i just go with P2200 or any GPU like 1660 or 1660s and jailbreak it with the custom driver? 

[Lessaj 198]

8 hours ago, Dalpuri said:

If i want the video to be transcoded almost instantly assuming I'll only have 1080p max videos. Should i just go with P2200 or any GPU like 1660 or 1660s and jailbreak it with the custom driver? 

I mean if you want them to be transcoded faster, because realistically it's not going to be almost instant anyway, then you would need to go with latest and greatest generation Ada cards. The P2000 is Pascal and the 1660 is Turing so it -should- have faster transcoding performance but I don't have any concrete evidence to support that. Buy the best card you can afford that fits in your chassis and you have power budget for, it's going to be a lot faster than CPU encode either way.

[Dalpuri 18]

13 hours ago, Lessaj said:

I mean if you want them to be transcoded faster, because realistically it's not going to be almost instant anyway, then you would need to go with latest and greatest generation Ada cards. The P2000 is Pascal and the 1660 is Turing so it -should- have faster transcoding performance but I don't have any concrete evidence to support that. Buy the best card you can afford that fits in your chassis and you have power budget for, it's going to be a lot faster than CPU encode either way.

This cleared my confusion thank you so much.