FWIW, I have found frigate to be extremely unfriendly to get to run unless your cameras are running wide open, do not require passwords, and you know how to code perfect YAML or whatever that is that Frigate requires.
As to your rig, yes that is overkill that will make the local electric company very happy. with any luck those processors also know how to reduce their power consumption as necessary. Good luck!
Look, maybe you’ll get lucky and the thing will work from day one as intended. I found the setup process to be extremely frustrating because any little error and the whole setup file is wiped. Other folk have managed to get their cameras recognized, etc. I am not there yet.
The only reason I am even considering putting up with this nonsense is to consolidate the IT infrastructure and better take advantage of Google Tensors. That said, if I do not succeed, Blue Iris running inside a Windows VM should work great on that humonguous CPU cluster you have.
To call it massive overkill would be a massive understatement.
You could support literally hundreds (or possibly thousands) of simultaneous NAS users with that CPU/memory spec. Or run some very heavyweight VMs and apps.
My advice would be to scale back the processor (to 1x 16-core) and memory (to 2x32GB) and use the money to to add a couple of SSDs for a mirrored pool to hold your apps and their active data (such as the Plex metadata).
I’d have to agree it’s overkill. I ran much more than what you listed on an e3-1275v6 quad-core from 2017 and typical utilization was 20%. RAM is always useful for cache, but if things aren’t repeatedly accessed before getting evicted then it won’t provide a ton of benefit.
I know what Frigate is for, but haven’t used it myself. I’d assume that like most other video-related things, you’d be much better off either buying a CPU with integrated graphics, or adding a discrete card. If the Arc GPUs have enough power for your needs, they support modern codecs, super cheap, well-supported on Linux, available in low-profile/half-height, and low wattage.
Did some more research on the available processors for the server I’m building out (Used server site TechMikeNY.com) and the lowest TDP package with the best thread count (because why not) is the Xeon Silver 4116 with an 85W TDP.
Everyone is saying how much it will cost to run, so I looked at one tool and it estimates (for the CPUs only) that as 25% load running 24x7 over a year would be $67 based on my electricity rates. I know the rest of the server will draw plenty of power, but thought this sounds way lower than I expected. Maybe the site I’m using is BS? Does anyone know of a good site for estimating power draw for a server based on components installed?
Very hard to know what the actual draw will be until the rig is running and the various tasks you give it are operational. Modern CPUs can be very good at throttling down. My gear here consumes about 1/4 the home energy budget and at $0.33/kWh, it’s enough where I am trying to consolidate blue iris into the server, again.
@Constantin Thanks! I figure using the lowest TDP processor should be my best bet at a “lower power” server. Fingers crossed I don’t blow up the power bill!
You don’t seem to have anything to do Plex Hadrware Transcoding.
Some Intel Arc graphics cards (A310 based) are low powered but very capable for this and come highly recommended
TDP is indicative of power use at full load. What matters is power use at idle, which would essentially be the same for all CPUs of the same generation irrespective of TDP.
However with Xeon Scalable you’re committing yourself to quite high idle power.
He didn’t indicate a need for transcoding either
And for occasional use, it might be easier to just do it on the CPU.
It appears to be based on running the CPU at its nominal TDP, which is not realistic.
A NAS idles most of the time. And if apps push your CPU to full load, you do need the power anyway.
In-between, you have to build and measure.
Motherboard/server reviews on ServeTheHome generally include some power figures, and may give you indications—taking into account that “idle” depends on what’s in the server, so a storage server with lots of spinning drives would use a lot more than the tested configuration.