Hardware for 4-5 4K and home lab - Printable Version +- Jellyfin Forum (https://forum.jellyfin.org) +-- Forum: Support (https://forum.jellyfin.org/f-support) +--- Forum: General Questions (https://forum.jellyfin.org/f-general-questions) +--- Thread: Hardware for 4-5 4K and home lab (/t-hardware-for-4-5-4k-and-home-lab) |
Hardware for 4-5 4K and home lab - Gabriel Freiberg - 2023-10-01 I’m looking for good hardware to support transcoding at most 5 x 4K streams and have a little juice leftover for other services I might run. This will be my only home server for now, and I will do some virtualization and other tasks Lower power consumption would be nice when idling. I’m honestly lost with all of the options out there. Sometimes newer processors list a huge amount of power consumption which is confusing. Here are a couple I’ve looked at HP EliteDesk 800 G6 Mini Core i7 10700T 2GHZ/16GB HP EliteDesk 800 G6 SFF, i7-10700, 2.90GHz, 8GB RAM, 256GB NVMe, Beelink Mini PC, SEI10 10th Gen Intel Core i5-1035G7 (4C/8T, up to 3.7GHz), 16G DDR4 500G PCIe3.0 SSD RE: Hardware for 4-5 4K and home lab - Venson - 2023-10-01 Hey there, For the core server, those options only with an embedded GPU look a bit underpowered. 5 x 4K streams are really too much for integrated gpus, i dont think you will get around a dedicated GPU here. RE: Hardware for 4-5 4K and home lab - bitmap - 2023-10-01 Look at the iGPU series for each and see how it performs. This short section of the HWA documentation is a good place to start. You probably want to avoid the "entry level" with the spec you mentioned: https://jellyfin.org/docs/general/administration/hardware-acceleration/intel#speed-and-quality. "A little left over" depends on what you want to run. My i7-13700k is probably more power consumption than you're looking for but can serve 17 containerized services, run either software or hardware accelerated encoding processes, still be fully functional via SSH, serve Jellyfin and direct play or transcode as needed all simultaneously without ever giving me a warning about system load. With htop/bpytop I don't think I've ever seen 1m/5m go above a 5...I feel like I'm trying to give this thing a run for its money and I can't. RE: Hardware for 4-5 4K and home lab - bitmap - 2023-10-01 (2023-10-01, 09:19 AM)Venson Wrote: Hey there, I really, really think a UHD 770 could handle 5x 4K HDR > 1080p SDR streams with proper config on the right platform (Linux). Likely even a UHD 750...might be cheaper in the short run to purchase something like an A380 (can't speak to how many streams it can handle) but the long-term power consumption of a dGPU would outweigh those upfront savings. I wish there was a benchmark document for Intel's iGPUs for transcoding. I can try to test the UHD 770 if you're interested...would be just me opening a bunch of windows and streaming 4K content on a machine that can't handle it. Might crash that machine but would give me some interesting data... If there's a better way to test and/or it would be helpful, let me know! RE: Hardware for 4-5 4K and home lab - TheDreadPirate - 2023-10-01 (2023-10-01, 09:38 AM)bitmap Wrote:(2023-10-01, 09:19 AM)Venson Wrote: Hey there, I have a UHD 730 and I am barely getting one 4K HDR > 1080 SDR stream. Tone mapping seems to be very taxing. I'm not sure the UHD 770 is 5 times more performant than the 730. IMO, I don't think you can get around having a dedicated GPU. The go-to is the Intel Arc A380. RE: Hardware for 4-5 4K and home lab - TheDreadPirate - 2023-10-01 I am going to amend my statement. It seems that my GPU is not being completely taxed for a single stream. But the conclusion is still the same. You can probably get away with 2-3 simultaneous 4K HDR transcodes on an iGPU. But for a better user experience you should have some headroom so I would still recommend an Intel Arc A380. RE: Hardware for 4-5 4K and home lab - bitmap - 2023-10-01 Quote:I have a UHD 730 and I am barely getting one 4K HDR > 1080 SDR stream. Tone mapping seems to be very taxing. I'm not sure the UHD 770 is 5 times more performant than the 730. The UHD 730 has a single MFX core is my understanding, while the 770 has two. I really don't know the best way to test performance on this front honestly. Probably not the right topic for that discussion. And we're just proving OP's point that all the options and lack of data out there are confusing and problematic. RE: Hardware for 4-5 4K and home lab - gado45 - 2023-10-22 Reviving this post Im also looking for hardware I need just one maximum two (at the rarest most) 4k hdr transcodes with tone mapping, would prefer igpu for the power savings, dgpu doesn't justify. My budget allows between i5 10th gen vs i5 11th gen, or go i3 12th gen. The server will be doing other home lab uses. What do you guys think? RE: Hardware for 4-5 4K and home lab - TheDreadPirate - 2023-10-22 You would probably still need a dedicated GPU. My i3-12100 can do 12 non-HDR transcodes. But can only handle 1 HDR tone mapped transcode. One of the devs stated that tone mapping requires a lot of graphics memory. Which is a problem because most motherboards don't allow that much memory to be assigned to your iGPU (512MB is the max for my board). Generally speaking it is recommended that you keep your 4K content in a separate library and pre-transcode all that content to 1080P SDR to put in your main library. And reserve the 4K HDR content for clients capable of direct playing them. RE: Hardware for 4-5 4K and home lab - gado45 - 2023-10-24 (2023-10-22, 04:55 PM)TheDreadPirate Wrote: You would probably still need a dedicated GPU. My i3-12100 can do 12 non-HDR transcodes. But can only handle 1 HDR tone mapped transcode. One of the devs stated that tone mapping requires a lot of graphics memory. Which is a problem because most motherboards don't allow that much memory to be assigned to your iGPU (512MB is the max for my board). Then there's no much benefit for spending for a newer gen cpu, I could save on older gen i7 and benefit from more cores, then pre-transcode or throw low watt gpu if needed. Or maybe even go ryzen for the build if no igpu will be needed. |