I’ve been running TrueNAS in various iterations for quite some time now, and am now running Scale 25.10 on aging repurposed hardware. I’d like to upgrade my motherboard/CPU to a modern low power solution. I generally only run the system during the day and have a scheduled power down overnight to save on energy. I’m also running a separate Proxmox machine on an N355 server which could make a great solution if I didn’t have an 8x18TB RAIDZ2 NAS (the limit on that box for eSATA is 4 drives, and that machine won’t take more than 16GB of RAM). I imagine some sort of Intel Core solution with all of the low power states enabled and enough RAM to support Apps and healthy ARC usage is probably where I’ll head, but the number of options makes my head spin. This is a homelab machine, mostly, so user loads are quite low.
Here’s the gist of my current system:
Base
Single Xeon E5-2640 on a Supermicro MBD-X9DRG-QF-B (I pulled the second CPU to save 35W)
112GB RAM
Dual 10GBe PCIe NIC
Intel ARC A310 (for transcoding)
Pools
Boot - 2x SSD Mirror
Apps - 2x SSD Mirror
Data - 8x Seagate 18TB spinny disks in RAIDZ2
Usage
Fileserver
SMB
Resilio Sync
Backup host
urbackup (host for remote servers)
“Data Protection” (push to cloud servers)
Apps
Media server (Jellyfin)
Immich
Various Docker apps
VMs
Legacy Windows VMs, spun up very infrequently for legacy access
In a perfect world I’d keep my BOXX case (ATX form-factor motherboard) and just upgrade mobo/CPU/RAM/boot + app pools (NVMe?) - pretty much only the ARC GPU and the 8x18TB drives are the only thing coming along for the ride. I’m curious about an N3xx system like this - is 48GB of RAM enough to handle that large of a RAIDZ2 and a handful of Apps? Current usage seems to be right around 48GB, but ARC naturally expands to fill what’s available over time.
I also welcome performance tuning tips that I could implement now/moving forward to shave watts. I originally had the spinny drives set to spin down on idle, but they kept spinning up every 5-10 minutes and that seemed like it wasn’t the best thing in the long run.
The official word from Intel, is that 16GBs is the maximum memory:
Many people have successfully used more memory on these smaller processors, (N100, N350, etc…). But, it is a risk for reliability.
Now as for larger RAID-Z2, any memory size will work. ZFS will just not be able to cache as much in RAM with less memory. Now for Apps, I can’t say, I don’t use the Apps you have listed.
Yes, well, ZFS was not originally intended to save power.
IMO these Intel Nxx “NAS” boards have only one thing going for them: Low power consumption.
Everything else, like stability and longevity, proven track record, IO possibilities (without port multipliers) and expandability (9 PCIe lanes…), not so much.
They are meant to consume little power and beeing cheap. Thats all.
8 SATA HDDs, 10GbE: You can run it all on an AsRock Rack E3C246D4U2-2T (new old stock on eBay from China) if boot and apps move to NVMe. Depending on cost of DDR4 UDIMM now; 128 GB is probably out of reach, but 32-64 GB may do.
A Ryzen motherboard with a SAS HBA.
Conversely, if boot and apps are on SATA SSDs, a Supermicro A2SDi-H-TF (-TP4F for SFP+) may do it all, using (lots of) (cheap) RDIMM. But you’re stretching this very low power board to its limits, and it won’t come cheap.
For a home server, I’d say that just the mirrored boot part was already overkill.
And the pairing of an expensive stupidly oversized “AI-hype” motherbord with a lower end CPU is puzzling.
If you want to ZFS-proof the build, mind ECC!
If you want to “future-proof” an all NVMe pool, go for EPYC 8000 with add-on NIC and GPU so you can upgrade.
Or already push up the onboard features, like this MSI D3052 board which has 25 GbE.
Edit Coming back on multiple M.2:
Xeon D-1500 and AMD EPYC 4000/Ryzen CPUs can do x4x4x4x4. So an AM4/AM5 motherboard could take 4 M.2 on a passive adapter in the PCIe slot. The MSI D3052, for instance, would still have a x4 slot on CPU lanes to host an Arc dGPU for QuickSync.
Intel Core/Xeon E and AMD Ryzen APUs can do x8x4x4. So an E3C246D4U2-2T could, for instance:
take 3 M.2 on a passive adapter in a PCIe slot; or
take an Arc dGPU for updated QuckSync in one PCIe slot and a pair of M.2 on a passive adapter in the other PCIe slot for a mirorred app pool; or
take an half-height Arc A310 and a pair of M.2 in a x8x4x4 riser in one single PCIe slot (useful for boards without a pair of bifurcated x16/x0-x8x8 slots).
You still have the on-board M.2 slot(s) on PCH lanes for (mirorred) boot. No need for an oversized E-ATX board and matching XL tower or deep rack.
Interest piqued. That post mentions “low power” but the thread doesn’t seem to back that up with any stats. At this point the lowest power option is going to be my favored solution, even if it means cutting a few corners. Leaning towards an i5-12500T (assuming I can find one) and ASUS Prime B760-PLUS D4 motherboard with PCIe addons. Thinking I might migrate my Proxmox setup over to this and run SCALE in a VM to save even more power. I already have a BOXX case which fits 8 drives and room space isn’t an issue, so I’d prefer to reuse.
Will still see if I can find any deets on idle power from that motherboard.
BMC only power off – 4.8w
Ubuntu 14.04 LTS installation screen – 31.3w
UnixBench 5.1.3 single thread max – 36.7w
Max observed – 109.1w
Common wisdom here is that T and non-T have similar idle power, and that ‘T’ may actually use more power under load because non-T ramp higher, complete the task faster and win the “race the idle”. So the main reason would be to limit maximal power if the case has cooling issues.
I’m unsure how well Prowmox behaves with P+E hybrid architecture. If you go this way, tough, you may as well go for a 13- or 14- ‘non-K’.
Fair points and nice recommendations. My current system is pulling ~150W at idle so generally anything less than that is a big win. While this is a “Home” server I still use it for work and critical backups, and I’d also rather pay for fault tolerance to minimize downtime, although I realize “fault tolerance” and “power sipping” are probably diametrically opposed.
If you want to keep lots of PCIe lands, the Intel X299 hardware is not too shabby. With an 8-core 9800X and six SSDs (SATA / NVMe mix) in an ATX build, I idle 50W - 60W.
