Chuck's Cool Reviews and Info

Projects, news and stuff going on. Pictures link is to the right.

Performance testing: Plex Hardware Transcoding with a Jasper Lake Intel Celeron N5105 CPU on Ubuntu Server

- Posted in Blog Entries by

In a previous post, I provided instructions for getting Plex Hardware Transcoding working on your 11th gen Jasper Lake Celeron CPU Server running Ubuntu.

The Aerofara system I am using here is outfitted with 8 gigs RAM and is running in dual channel mode.

For this one, let's talk about the hardware transcoding capabilities of this processor vs. 4k content. Frankly, it blew me away when I first investigated it. See images below to see it powering through some top shelf 4k bitrates.

I want to show repeatable scenarios for you to test with your systems, so I am using the Jellyfish bitrate test files as test dummies for this effort. You can download them all for you own testing here: https://jell.yfish.us/

I created a Video Test Files library in Plex and added all of these files to it. I then played every single file. They all played, and the 4k ones hardware transcoded down to 1080p, as that is my monitor's resolution. I was amazed. This is a $220 crap computer. My previous crappy Plex server running on a Celeron N4100 struggled with anything 4k.

Then I wanted to know how far I could push it. I put the largest video in this collection (jellyfish-400-mbps-4k-uhd-hevc-10bit.mkv) into a repeating play loop and got two streams going before it fell over with excessive stuttering and buffering.
This was playing over a wired connection and my destination computer had plenty of power remaining. I am assuming that the integrated GPU was at its max. The CPU was doing fine with plenty of available capacity as well as plenty of available system RAM.

This is a rough test, as these are 30 second clips set on a loop, so it gets going and then has to stop and start again every 30 seconds. It is maybe possible that with a longer video that this CPU could do more hardware transcoded streams as it had more time to compute and buffer more streams rather than starting over every 30 seconds.

enter image description here

enter image description here