Hi, I’m new here and was looking for general advise, I’m looking to replace i5 6500 + 1060 6G for a Intel Ultra 7 265K in my homelab, my rational and use case is mainly jellyfin server for couple of users and for that the iGPU is well enough. I got immich, frigate, home assist, etc. I plan to have minecraft and other games servers in the future.
Another point is i5 6500 is never at idle so the power draw is bit higher.
Does this upgrade make sense?
Can’t speak to all of those services as I am not aware but I am able to get away with 4K transcode on a 11th gen i5 iGPU, but this is only one user. As long as the host is wired the buffering is minimal.
If you go with a discrete GPU I am pretty sure there is a way to dice what services use what to divvy up the workloads… I.e. only use this GPU for FFmpeg. Home assist can run on a pi so I don’t think that will make a dent.
If you have a ton of users that won’t be taking advantage of direct play on Jellyfin you may need some help on the GPU side.
Heres what I am kinda doing here with 11th gen i5
Jellyfin server; multi user, with iGPU one transcode for 4k (I use something like blade runner as a proxy since its so intensive). Direct Play for everyone else [or I kick them off 
]
Airsonic Server; multi user with transcoding and active podcast management
Nextcloud server, I never used immich but I think its comparable?; multi user, picture and document traffic, client auto uploads pictures/videos from like an iPhone or something.
Scheduled Backups of user home folder weekly, using compression.
Isolated VMs with 1-2 threads each (1 - 2 at a time).
Gaming like bg3 but that workload is for the 3080 TI.
I am definitely pushing what I have a bit too much but if this gives you a picture of what you can do with the multi-generation leap I hope it helps.
Unfortunately I can’t speak to hosting game servers or frigate.
That’s my pointless non-answer answer for the day.
WWED
Thanks, that’s actually helpful to know what you can get away using the iGPU, gonna do a follow up if I get to build it.
The igpu option is fantastic, but you can get a ton of transcoding performance out of jsut an arc a310. I can get about ~90FPS transcode 4k to 4k, 10 bit HDR to 8 bit SDR with tonemapping, so really its good for 3 high quality streams. Going down in resolution to 1080p is well into triple digits fps.
If you dont need the lanes for HBAs or other cards, its hard to argue with the price/perf of arc cards.
I have intel in both of my servers because I like to be able to pin less important, but more active workloads to the E cores. This makes a significant difference in what my UPS says its run time is, though I havent measured the real difference in draw. Its mostly so I can reduce heat in my unconditioned closet.
Another route could be breaking up your jellyfin server to a dedicated n100 minipc. The N100 is basically around haswell-skylake performance for very little power draw but more importantly it has a modern iGPU capable of 4k transcodes with tonemapping. It definitely punches well above its weight for this task and they can be found for pretty cheap comparitively.
I bring these options up because the 265k by itself isnt the most compelling buy in the world. Not to talk you out of buying it, but it is a decent investment into a modern platform with meh performance for the amount you would be spending on it.
I think it would make sense from the power perspective, unless you have super cheap power. It probably is an overspecced system just for Jellyfin, immich, etc. don’t know how much the game servers need.
If you do get more users you can always add a dGPU later.
Welcome to the forum!
I think I find myself in agreement with Adubs here. The core ultra 7 265k is kinda expensive for your workload. I’d definitely split the builds. HomeAssistant can run on an odroid n2+ (in fact that’s used in their own home assistant blue variant, which was discontinued in favor of yellow, a rpi cm4 variant, but I digress). Also, the 265k has a massive power consumption compared to the 6500. Even if the 265k will be idling, it’ll likely use about as much power as the 6500 now.
I never ran frigate. But looking at it, the system requirements are minimal (intel n100). That’s where the might all-purpose odroid h4+ or h4 ultra comes into play. You can get either an intel n97 (which is actually newer and better than the n100 - thanks intel for your stupid naming convention) or the i3 n305 (basically an 8 core variant of the n97). I believe this should be capable enough for homeassistant and frigate.
Keep the old i5 6500 around, but remove the GPU and use the iGPU for jellyfin. I feel like most of your power draw is from the dGPU and if you can use the igpu for the encoding, you should be in a better spot. Or you could just buy 2x odroid h4+ boxes and have a theoretical total power draw of maybe 30W (excluding the storage).
Immich should be peanuts basically. Given the ultra 7 265k’s $400 price tag (just for the CPU) you can easily buy 2x odroid h4+ that will use less power for the job. Now the question then becomes: “do you have a switch already?” If you do, then perfect. If not, you’d probably still end up cheaper with 2 builds instead of one larger one.
For the lower power draw and for the price of the h4+, I find it hard to justify even going to a lower ultra 5 245k. The only disadvantage of the odroids is the lack of pci-e expansion. If you want to add stuff later, like a gpu, well… tough luck, because the thing only has an m.2 slot. But if you plan accordingly, you should be fine.
Bonus for split builds: you can keep the frigate + homass build 24/7 on, but be able to shutdown your jellyfin system to save some power while it’s not in use. You can then use the homass system to WoL the jellyfin system (assuming you don’t mind waiting maybe about 2 minutes for it to come online to save some power). But given that you’re running more than just jellyfin (you mentioned immich and minecraft), you’d probably be keeping the system online (which is still fine, the power draw is just so low).
The n97 is about 10-15% faster than the i5 6500, while using 1/4 of the power. I would’ve recommended the h4 ultra more, because the n305 is basically double the performance, but idk if you can use the igpu of the n305 for both frigate and jellyfin at the same time. And you said you have “a couple of users,” which means the GPU might’ve been a bottleneck (if you only have one for all your workloads). So in this case, 2x h4+ make more sense than 1x h4 ultra.
I still have to highlight that it really depends what you’re planning on doing in the future too. You could technically get something like the ultra 5 245k and add an intel arc a310 to power both your jellyfin and frigate systems (well, in theory you could jerry-rig an a310 on the h4 ultra, but that’d be too jank even for my standards - unless you get the h4 itx case and slap the adapter inside, somehow, without shorting something on the back, which should be doable and less jank).
But if you want to run more stuff in the future, then you might want to get a proper “full desktop” build (245k / 265k / ryzen 9600x etc.).
So it’s your turn to pick your poison. I personally really wish we’d see desktop systems that use less power.
Love the feedback, looks like n100/n97 would be the logical choice for a simple transition.
I should have started with more context on the current setup.
8-port 1Gb TP-link switch
Rpi-4 running nginx and Home assist
Truenas server running on:
i5 6500
1060 6G
48GB DDR4
3 x 14 TB Seagate RaidZ1
1 240 GB nvme cache
Power draw with 1 jellyfin directplay stream
Future plans/wishlist
If there’s a specific test somebody wants run, I’ve got a H4 Ultra (N305) and one of those Asrock N100 boards I could run the workload on. Just would need to know how to set the test up.
The one caution I’ll give about the H4/+/Ultra is that the mini ITX kit is not perhaps as all-inclusive as I might like. It’s a minor thing, but it’s also held me up for a few days now – converting from normal 9-pin front panel IO into the H4’s GPIO layout is something one will have to sort out if they want to mount it in a normal case. It’s not ‘plug in and go’, in that regard.
I believe you would need to set up the i915 driver and then sr-iov should be possible on the igpu. I don’t know what all is Involved personally but I know it’s possible.
Should be the same process.
It looks like he was using an older kernel version. You’ll want something 6.8+ for driver support of the n100 iirc.
Sorry for the late reply but wanted to do a follow up on the homelab build using arrow lake with TrueNAS.
So far no iGPU in ElectricEel-24.10.1
ls -alh /dev/dri
shows no devices
so does
intel_gpu_top
Not even gonna try messing with the NPU
Gave it a shot with a N305 on an Odroid H4 Ultra on the 1.06 bios following the steps from that and the github repo it’s pulling from:
The 6.8 steps do work on this device, but one will have to pin that specific kernel version (6.8.8-2) as Proxmox is up to at least 6.8.12 after the post-setup update install.
