Prospective 2U Rackmount Proxmox/ZFS NAS Build – Looking for Feedback

[Mancolt]

Hi everyone,


I’m planning my first serious rackmount DIY NAS/server (I've built many desktops, but never a NAS or rackmount server!) and would love some experienced eyes on the build before I pull the trigger. I've been mostly building it with the help of GenAI models, and I've noticed some issues in those suggestions. Initially providing an AMD mobo and an Intel i3/i5, and suggesting I buy a Titanium rated PSU, when the chassis we're looking at includes a PSU (albeit gold rated, but it would take decades to see a positive ROI due to the slightly better energy efficiency). So I don't have complete confidence in the AI recs.

Goal is a reliable, power-efficient Proxmox + ZFS storage appliance (storage-first, light app/VM use) that will live in a Sysrack 27U (https://www.amazon.com/Sysracks-Locking-Enclosure-Networking-Computer/dp/B0B4DZGW4B/) in my basement AV rack. Started with 2U, but have plenty of space for a 3U or even 4U build here. My 27U rack has maybe 10 of the 27U allocated so far to my Unifi and A/V equipment, and I don't envision ever needing all 27U of space, but I had a cabinet that would fit the 27U perfectly, so here we are.


Use case

• Primary: Plex media library, Immich photos/videos, important files, audio, and UniFi Protect event archiving (likely only key clips via Off-Site Archiving — NOT full 24/7 footage duplication).
• Secondary: Frigate (live AI detection on camera RTSP streams, not re-processing recorded footage), light LXCs/VMs.
• Existing setup: UniFi XG 8 PoE (needs dual 10GbE RJ45), UNVR with 4× WD Ultrastar DC HC570 22 TB (5th drive is cold spare), smaller 12TB QNAP for critical-item backups only, future 3-2-1 offsite NAS.
• Priorities: Reliability (ECC, IPMI, ZFS), good value, low-ish power (~80–110 W idle target), short-depth 2U rack fit, dual 10GbE RJ45

Proposed Build:

• Chassis: Supermicro CSE-825TQ-563LPB (short-depth 2U 8-bay, includes 560 W 80+ Gold PSU) → Supermicro SuperChassis SC825TQ-563LPB – RackmountNet
• Motherboard: Supermicro H13SAE-MF (AM5, IPMI, full ECC, 2× native M.2 PCIe 5.0) → Supermicro H13SAE-MF Motherboard Micro-ATX AMD EPYC™ 4005/4004 series, Ryzen™ 9000 & 8700G* & 7000 Series single socket LGA-1718. - Newegg.com (Manuals: https://www.supermicro.com/manuals/motherboard/H13/MNL-2627.pdf)
• CPU: AMD Ryzen 5 8600G (65 W, Radeon 760M iGPU for Frigate/Plex VA-API) → Amazon.com
• RAM: 128 GB ECC DDR5 (4×32 GB server-grade kit, validated for board)
• Networking: Intel X550-T2 dual 10GbE RJ45 PCIe card (for XG 8 PoE)
• Boot pool: 2× WD Black SN850X 500 GB NVMe (mirrored)
• Cache/Special vdev: 1× WD Black SN850X 1 TB NVMe + low-profile PCIe x4 M.2 adapter (Vantec UGT-M2PC130)
• Storage pool: 6× Seagate Exos ST22000NM000C 22 TB recertified (matches UNVR spare specs for shared cold spare) → Seagate Exos ST22000NM000C 22TB 7.2K RPM SATA 6Gb/s 3.5in Recertified Hard Drive
• Planned config: Single 6-drive RAIDZ2 vdev (~82–86 TB usable), Proxmox 9, ZFS compression, small dedicated datasets for Plex/Immich/Frigate/Protect events.

Questions for the community:

1. Any glaring compatibility issues or known gotchas with this combo (especially H13SAE-MF + Ryzen 8600G + X550-T2 in Proxmox)?
2. Better value / higher-reliability alternatives in any category (motherboard, CPU, 10GbE card, NVMe, etc.) even if it costs a bit more?
3. Is the included Gold PSU fine, or is upgrading to Titanium worth it at ~80–110 W idle?
4. Any suggestions for lowering power draw or improving Frigate/Plex performance on the 8600G iGPU?
5. Overall value — am I overpaying anywhere or missing a smarter way to hit the same goals? I'm not trying to cut price, but don't want to pay for performance that will never be used, or for components that are a poor value (swallowing the prices of the NVMe and HDDs right now feels like that, but this is new reality and I don't want to wait years to do this build just to see if prices ever come back down).

Thanks in advance for any feedback — I really appreciate the help before I start ordering parts!

 

[alaricljs]

Your special vdev needs redundancy. If it dies you lose sll your files. It's not likely that your workload needs L2ARC if that's what you mean by 'cache'

 

[louie1961]

I don't think that CPU is officially supported by that board, and I don't believe that CPU will support ECC memory. I would go with an AMD Epyc 4545P and a separate video card. It will make passing through the GPU to docker apps much easier. On a side note, I am not a big fan of trying to combine a NAS and a server in the same box. I have my Proxmox server (Ryzen 7 5800XT/Asrock Rack X570-D4U/128GB of ECC RAM) separate from my TrueNAS NAS box (Asrock Rack D1541D4U-2T8R/Xeon D -1541/64GB of ECC RAM). I think TrueNAS does a better job of data protection and backups than Proxmox does. I also run Proxmox Backup Server as a container on my TrueNAS box, so if I need to rebuild Proxmox, my PBS isn't living on the box to be rebuilt.

A couple of other thoughts: why are you going with ECC RAM, but using consumer drives? Seems like a bit of a contradiction. Especially in a special VDEV. You really ought to have enterprise class drives with power loss protection in a special VDEV, and they should be mirrored for protection. If you lose your special VDEV, you will lose all of your data. A raidz2 main pool is not going be very performant for your VMs. In fact it is about the slowest configuration you can choose. My Proxmox server has four smaller sm863 drives in two mirrored vdevs as my main pool and it is pretty fast. I keep virtually no data on my Proxmox node, so I can get by with a 2GB pool. That makes my VMs almost ephemeral. My TrueNAS box on the other hand has six drives in three mirrored vdevs, plus a pair of optane devices as a SLOG and a pair of Micron 7300 Pro NVME drives as a special metadata vdev with small blocks. That pool is plenty fast, even over NFS. It will easily saturate my 10gbe network. All my data lives there and backs up to my synology. Finally, what does your network look like? You are specifying a 10gbe NIC, but with rj45 connectors. If performance matters, SFP+ links have lower heat and lower latency.

 

[Mancolt]

Thank you both for the detailed feedback — I really appreciate it! I've still got a ton to learn on this front, so much of this is over my head. I know enough to know that AI is making some obvious mistakes on its suggestions, so I can only imagine there are mistakes on some of the more nuanced elements.

I dont want to give up 3 drives worth of capacity in a 6 drive setup. None of my stuff is that mission critical. What little there is will be backed up to the QNAP and eventually a 3rd NAS offsite.

I'm still evaluating, but im leaning more towards Unraid than TrueNAS just for ease of use/lower learning curve.

@alaricljs — Great catch on the special vdev. I was planning to use it for metadata/small-block acceleration on the 6-drive pool. I’ll either mirror it or drop the dedicated special vdev entirely for now.
@louie1961 — This is the feedback I was hoping for. I think I’m also going to follow your advice and split the build into two separate boxes:
Dedicated storage NAS (Unraid or TrueNAS Scale) with the 6× 22 TB drives, 64 GB ECC, simpler AM5 board, and single 10GbE.
Separate Proxmox compute box (likely a used/refurb Dell R650 or HPE DL380 Gen10/11 2U) for Plex, Frigate, Immich, and any future LXCs/VMs.
This gives me better isolation and makes it easier if I ever grow beyond light services. It also just makes more intuitive sense to me and seems like ill have an easier time managing things on separate, purpose built devices. I’ll keep the NAS new/scaled-down for reliability and probably go used-enterprise on the Proxmox side for value.
A couple quick follow-up questions if you have time:
For the Proxmox box, would you lean toward an R650, DL380 Gen10/11, or a Supermicro SYS- series in the $900–$1,600 used range?
On networking — since the Proxmox box will be doing most of the heavy lifting to the NAS, is dual 10GbE on the Proxmox box + single 10GbE on the NAS the right way to wire it to my UniFi XG 8 PoE?

I have the Pro XG 8 PoE in my rack. On the opposite side of the house, where all the legacy Cat5e runs terminate, i have a Pro Max 16 PoE. Im hoping to pull a Cat6a from my office to the Pro XG 8 PoE to bypass the Cat5e/Pro Max 16 PoE bottlenecks. My current plan is to connect my UCG Fiber to Pro XG 8 PoE via short DAC to SFP+ ports (both in the 27u rack), and then connect the XG 8 PoE to the Pro Max 16 PoE via a 40ft OM3 fiber run, using the other SFP+ port on the Pro XG 8 PoE. So I dont have any extra SFP+ ports, and the next option in the unifi lineup with more sfp+ ports is the Pro XG 24 PoE, but that's $1800 and not something I really want to spend on unless necessary. Thats why I was looking at RJ45. If im better off connecting the Pro Max 16 PoE and Pro XG 8 via RJ45, freeing up the SFP+ for either proxmox server or DIY NAS, then I can make that change. But if you had to choose between a single SFP+ connection or dual 10GbE RJ45, which would you pick?

Thanks again — I just recently found this community but have already gotten tremendous value out of it.

 

[louie1961]

I am not a huge fan of most used enterprise gear. Its often very loud and energy inefficient. I think you were on the right track with the Supermicro H13SAE-MF motherboard and the Epyc CPU I mentioned earlier. That will give you all the compute horsepower you will need for a long time. NAS boxes need significantly less CPU power. This is the place where you can get by with used enterprise gear like a used Xeon D SoC motherboard. To build my NAS, I purchased a used Datto S3P2000 1U server on Ebay for $250. It came with a Xeon D-1541 Asrock board, with IPMI, and 64 gb of ECC memory. It also has 14 SATA ports. I pulled it out of the 1U case, and popped it into a 2U case and now its nice and quiet. I added two six bay icy dock adapters to my drives, since they are all SATA SSDs. If you need more room for drives, put it in a 3U or a 4U case.

I like to have two connections on my server and my NAS. I dedicate one NIC to the management interface and all of the client traffic. I dedicate the second NIC for a non-routed storage only VLAN. Or if you have only the two boxes, directly connect those second NICs with static IP addresses. This way there is no network congestions/traffic/contention for the storage traffic. If I had to choose between 1 SFP+ or two RJ-45 I guess I would go with two RF 45 connections.

I am not familiar with the Unifi networking gear, I don't use it here. But generally speaking I have a mixed environment of SFP+ and RJ-45 connections. I want all of the stuff in my rack to communicate over SFP+ for the reasons I stated earlier: Lower latency, less power usage, less heat. My clients all connect by RJ-45 or wifi. But I have a very flat network. I can fit everything into a 16 port switch. My QNAP switch is 8 RJ45 ports and 8 SFP+ ports. My wifi AP is a TPlink EAP670, my router/firewall is a four port fanless N100 device running pfSense. Neither of my wan connections are very fast, like 600mbpsor so, so I don't need 10gbe at the router. I make sure I minimize my inter vlan routing with my dedicated storage VLAN that only exists on my switch, its not even included in my pfSense instance.

 

[TrashMaster]

Hunt around on facebook marketplace or ebay or craigslist or ... whatever local market options exist. u can often get used very nice rack enclosures for a couple hundred bucks.
i have had great luck swapping server fans out of noisy dell/crisco/stupidmicro gear and replacing them with arctic server variants that are way quieter. good luck