So I’ve been googling to find out what to buy for a jellyfin/game server and the problem is almost all posts are about buying hardware in the US or EU. I checked the second hand market and it SUCKS around here, like 10 year old+ CPUs being sold as ‘intel i7’ for stupid prices.
I’ve decided to buy new stuff, at least I get warranty. When I searched around for info on that I also got a lot of useless (for me) advice like buying mini pcs that are either not available around here or very expensive. Basically I think I just need to build something myself using non server hardware, is that a bad idea?
What I have in mind right now, please tell me if there is anything that will cause issues later:
-
i5-12400 (mainly for the integrated graphics to use jellyfin’s hardware transcoding. Is the UHD 730 enough for that?)
-
No idea what to get for MB, I usually buy AMD. H610 boards seem cheap enough but very limited on storage options (sometimes only a single pcie slot even), I do want to eventually add lots of storage to it.
-
Server quality storage is extremely expensive, not sure if its worth buying. Was think about getting some WD blue 2tb drives, but really not sure here either.
-
For case I saw some suggestion for cases with a ton of hotswappable drive bays but sadly I couldnt find any to buy in my country, will probably get whatever cheap one I can find.
-
PSU I want to get any gold PSU with low wattage from a reputable brand but its been hard to find 500W or lower here, usually shops only offer brand PSUs that are 650W+.
-
16gb of any decent ram I suppose?
Btw I live in Brazil, if you feel like searching around even tho the websites are in portuguese:
www.kabum.com.br
www.pichau.com.br
www.terabyteshop.com.br
Any help is appreciated, its gonna be my first server so I have no idea what I’m doing
Let me expand on that.
there is a really nice wiki article about intel hw encoding with color table https://en.wikipedia.org/wiki/Intel_Quick_Sync_Video#Hardware_decoding_and_encoding
1.1. the color does not show in dark mode on mobile
1.2. Your CPU is under the Alder Lake, so HEVC 12b encode/decode and AV1 decoding. This means you can encode big things into HEVC 12b to save space and also play back AV1 without issues in the future.
my setup and experience
2.1. I have an i5-8400 so I encode everything too big into HEVC 10b. I sometimes watch with 4 other people, so i had to tweak the config a lot to make it work fine when 5 people are watching at once. From my experience, the jelly sometimes just puts everything into decoding the first stream while others just wait. I recommend just setting only 1 core per stream and only like 300s transcoder buffer.
2.2. playback devices do a lot, when using firefox, everything has to be transcoded except h264
on chrome, it usually just remuxes (mkv container is converted into something more compatible) easy on resources
when playing using a jellyfin client, it is played directly
dedicated GPU
3.1. I disagree with the dedicated GPU recommendation as long as you tweak the settings, you’ll be fine.
Test iGPU first, play around with it, see how many streams at once you actually have to serve.
The dedicated GPU is an extra investment for a start and a constant power consumer for later.
yeah, dont think the dedicated GPU will be necessary, most posts Ive seen clain that intel’s igpu can handle multiple 4k streams, tho I could try to find a cpu with the 770 instead of the 730? Not sure how much of a diff it would make tho
My guess is the 770 would be a bit better. Could not find anything that could compare the two based on video performance and not strictly on gaming/rendering. So my opinion is not to care about it too much.
Yeah, might do just that then, thanks
I’m with you that he doesn’t strictly need a gpu, but if the price is right (free from old gaming PC, cheap from a friend’s old gaming PC, cheap old workstation card, etc) I stand by that he probably wants one. A lot less fussy, a lot more capable, nad nvenc does better quality encoding at lower bitrates (and probably less power too if you take into account time spent encoding at full tilt.)