Looking to replace my r730 hyperconverged host with a lowish power cluster

I currently have an r730xd LFF with 4 drive midplane with dual 2690V4 cpu’s, and 256 gb of ecc ram, and 14 x 16tb spinning disks, plus some nvme and ssd. I know the spinning disk eats up a lot of power. It’s all one proxmox and storage host. I’m hopeing to wind up with a similar power envelope but add some performance/redundancy.

I host a lot of services with docker under an LXC and a few with VM. The big ones are: *Arrs + Plex, frigate and blue iris NVR, Home Assistant, Microsoft Exchange (email), a couple virtual desktops, Nextcloud (back up all my data), Immich (back up all my family photos), FreePBX (important i have a small business and use it for my phone system), virtualized router with vyos, i have 2gb/2gb FTTH with a /29, and my isp is coming out with 5gbps soon. There are a bunch of other services I host, but those are the resource heavier one.

What i’m thinking of doing is:

I have an old 16 bay 3-u super micro chassis I’ll probably pull out the redundant PSU as well, and hack something more efficient in there. IPMI would be nice. I’m thinking a lower power enterprise board that supports registered ecc memory is important for my nas. I would also host my apps that i want to have ECC for, specifically nextcloud and immich. So i do need a little bit of power to run a VM and/or docker containers. I may try truenas scale, but i may just stick with proxmox so i can do nextcloud and immich in an lxc container, and run Proxmox Backup Server on my nas. So looking for suggestions for a board/cpu that can idle with very low power, but spin up if necessary, yet supports registered ecc memory… Any ideas?

Next I build a 3 node proxmox cluster with some lowish power PC’s. The new minisforum ms-01 seems like the best option? I’m looking for alternative suggestions. I’d put 96gb of ram in each node. I could use the thunderbolt ports to create a 3 node ring for ceph, which if it works should run at alteast 20gbps which should be good enough. That leaves the 10gbps sfp+ ports to uplink to my switch for data. I could use the 3x2 m.2 slot for a google coral for frigate. and the 4x4 and 3x4 nvme slots for a couple of 4tb nvme’s, my concern can i run heatsinks on the nvme’s in these slots? Esp since i think all 4tb nvme’s are double sided? My hope is to use quicksync for frigate & blue iris for encode/decode, likewise for plex. Frigate and Plex would be hosted in an LXC so it’s no problem to share, blue iris being in a VM i could pin it to a different node. But i could always put a small gpu in the pcie slot if i had to. I imagine that using the pcie slot will throttle the CPU due to power/thermal budget. Exchange would wind up running here, which that’s the one thing i’m a little bothered by no ECC. But i have 3 other clustered exchange hosts in different geographic regions.

so in summary:
1.) I’m looking for suggestions for a lowish power board/cpu that supports registered ECC for the nas that can also have enough power to run a couple VM’s.
2.) looking for confirmation that the MS-01 is a good choice for a proxmox cluster, or get some alternatives.

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.