Is Quick Sync Video any good? Media Server Hardware Encoding Tests | Enquire within, help needed!
ฝัง
- เผยแพร่เมื่อ 20 ก.ย. 2024
- If you're building an always on system which is doing any kind of video transcoding then you'll want to investigate Intel Quick Sync Video. Many common video applications can take advantage of this hardware transcoding technology such as Jellyfin, Plex, Frigate, Blue Iris, and many more.
However, we don't really have any standardised testing to allow us to benchmark iGPUs for transcoding performance. There are plenty of benchmarks out there for gaming performance but nothing for those us running these types of always-on server workloads where energy usage matters just as much, if not more, than ultimate performance.
Therefore I'm pleased to announce the first release of a Quick Sync benchmarking script allowing us to do just that.
A big thank you to my buddy Morgan Peterman (aka cptmorgan-rh) who wrote the majority of this script.
Blog post - blog.ktz.me/i-...
Github repo - github.com/iro...
Results page - gist.github.co...
Sanitzed results - blog.ktz.me/th...
===
🖥️ alex.ktz.me
🎙️ selfhosted.show podcast
🦣 mastondon techhub.social/@ironicbadger
📓 blog.ktz.me
💾 perfectmediase...
🧑🏽💻 github.com/iro...
🎹 Intro Music - Joe Ressington
The results are now published!
blog.ktz.me/the-best-media-server-cpu-in-the-world/
Such a great project, thanks for all who contributed/ will contribute to this project 🙏
I have Intel 5 6600k in z290 mb.
I’m running proxmox as host os, however, I’m not sure how pass through/ share the igpu to jellyfin lxc container
I assume that I’d need to install docker inside Ubuntu 22.4 then get jellyfin inside the docket!
The issue is how to map the igpu to the docker!
I’ll look for a tutorial and try to upload some results
See the linked blog post in the description, it has full details on running the container.
Nice project! I currently run my Unraid server on a MSI Z590 MB with an i5-11600. When I was shopping for new HW a couple years ago HW transcoding together with decent power draw was my number one priority. I started looking at i3 cpus but the iGPU in those were always much worse than the i5’s from the same generation. Tried to get an i5-11500 first, but found a deal on the 11600 so bought that instead.
I’ve been happy with the performance of the Intel UHD 750 iGPU so far. I tried HW transcoding a remux 4K HVEC 10 bit video down to 720p h.264 8-bit with HDR tone mapping. I manage to do at least 10 streams at once, then I ran out of devices to test with and was so blown away so I didn’t push it further. It’s just fantastic what it can do, without breaking a sweat.
Haven’t done much testing on power draw, and I really want to dive deeper into the BIOS and maybe underclock. Fiddle with P-states as well. Running 12-15 HDDs draws some power as well, even though I try to keep them spun down most of the time.
But, I actually had to RMA my CPU a few weeks back (thankful I bought the boxed version with 3 yrs warranty). I started having very weird problems with checksum, SHA256 library and crypto library errors. But Intel swapped it within a few days, so I’m back to happy. Never had a CPU go bad like that though, so it was a learning experience.
If I were to redo my system today I’d actually go away from the “one big server to rule my homelab”. The situation with the failing CPU got me thinking. I’d run the HDDs on a very low powered system, but it must have at least so many PCIe lanes that I can use an HBA card.
Then I’d run a hypervisor on a mini PC/NUC with built in Intel Iris Xe graphics. That iGPU is basically an ARC on a mobile chip. In every test I seen it has outperformed every desktop iGPU. So in a very small formfactor, with very little power draw, I’d get a HW transcoding monster for Plex.. 😅
Then for more compute intensive tasks I’d get a third “node” that can be designed for whatever tasks I need.
Yes, it’s nice to console everything into one system for some. But the redundancy in having spread out the tasks over multiple machines might very well be worth it. And probably not with any more power draw than an older system, likely less. The new P and E cores seems very efficient.
I’ll try and give your script a run, and put a Shelly plug in to measure power draw. 👍
So where is this upcoming video?
Love your content!
Love it! Last upgrade cycle my research pointed to the i5-XX500 as the sweet spot since it has the UHD770 and 2 “multi-format codec engines”. I did not end up getting 2x the frame rate in Tdarr though. I’ll add my systems and look forward to the results!
Appreciate it! It stands to reason that the 10th/11th gen should be about the best sweet spot. But that’s just my instinct talking.
The early results are showing not a lot of difference on the h264 encoders between generations or even clock speeds. Only if it’s a mobile or some other energy usage focused chip do the speeds really take a hit.
One thing holding true universally so far is that 10bit 4k hevc is a beast!
Future video idea: How do you manage updates/docker images across all your home devices? I have 3-4 physical linux bare-metal boxes, 4 pi's, and a multitude of other gear. Currently, I just ssh into each device once a week and run an update script to pull os-level updates and docker image updates. It would be very cool to have a Grafana (or like) pi that watches for updates across select hosts and displays "smokeping image on Pi2 is out of date", shows main server's HDDs and temps, the opensense info, things like that. I know there's tools out there like watchtower, Diun, portainer, others but have never really used. Would love to see how you manage this.
I have a 13th gen in my media server and it's great for transcoding. Since the CPU is mostly idle because of this I can use it for other tasks such as game servers.
If you’re able, please run the benchmark! Getting the newer CPUs on the charts is super important!! 🎉
Did you ever publish the results?
I am super interested!!!
Operating:
ASRock E3C226D2I, E3-1271v3, 16GB ECC, SF450 Gold, RTX3060, 3x 8TB HDD, 500GB SSD and it draws WAY TOO MUCH power.
Need help to improve it for sure!
Looking forward to seeing the results! I am running an i7 7700k for my unraid server and also have the itch to upgrade.
Probably won’t be worth it! Based on what I’m seeing.
Will the benchmark also test encoding performance?
is there a follow up video? the github shows, that 4k hevc will not even playback with a newer i5, is that correct? crazy to think that
Got something in the works soon(tm)
The results are now live!
blog.ktz.me/the-best-media-server-cpu-in-the-world/
Does things like direct or NAS storage come into play? For instance I’m running a Lenovo Tiny PC with Plex but the media is on my NAS
Shouldn’t matter for the sake of these tests. 😊
@@AlexKretzschmar awesome thanks
Wish I could help, I have an i7-9700 but I am running Windows.
NB while amd cpus may have more pcie lanes, the motherboards rarely manage to have more pcie slots.
That means MOAR NVME!
@@ktzsystems lol true
Which cpu will generate most quicksync fps per watt?
I installed a fresh copy of Proxmox 8.0.3 and attempted to go through the instructions as root. However, the docker binary does not appear to be found. Was there something else you did on your Proxmox system? Should this not be run as root?
Docker is a dependency.
docs.docker.com/get-docker/
Hmm doesn’t the apt get install docker line in the instructions satisfy that dependency?
Highly doubt my server has quick sync, old dell r310. I might build something but not using a media server right now, anyway the kittens pulled the only monitor to the ground. Our TV died before also
Those older Xeons were a good value a few years back. Doesn’t look like this one has quick sync though.
You are exactly who this video and research is for. The “how much am I missing out on” folk who want to replace an old CPU system that may not be the most energy efficient but has enough CPU grunt to get by.
But then when you factor in the energy usage of cpu transcoding vs hardware, the picture becomes less clear.
If anyone can figure out an vaapi based test command i can submit my 5700G and 4600u, as he mentions going AMD.
i tried swapping h264_qsv for h264_vaapi but i get the below error when testing on my working jellyfin lxc container...
>>Impossible to convert between the formats supported by the filter 'Parsed_null_0' and the filter 'auto_scale_0'
>>Error reinitializing filters!
>>Failed to inject frame into filter network: Function not implemented
Support is available on the self-hosted podcast discord at selfhosted.show/discord in the #perfectmediaserver channel using the ‘qsv-testing’ thread there.
Where’s the result to this comparison?
Working on it :)
@@ktzsystems any ETA? :)
@@Jodez23 blog.ktz.me/the-best-media-server-cpu-in-the-world/
Not sure what the issue is but no sound... 😢😢😢
Works ok here? Still broken for you or not? No idea how that’s even possible!
Try restarting YT
Wfm. Agree try a restart
@@ktzsystems I think it was the Amazon stream version - it is fine on my Mac today but via the Fire TV stick nothing.
I've seen odd things with Amazon before - I think they cache and play around with TH-cam but never bothered to Wireshark it TBH...