I’ve documented two builds of this server before, which I call Snuts: the NK4 Node 804 build and the 15 bay L4500 DAS/OTiS hybrid that followed it. That little X9SCM-F served me faithfully for years but a few things were coming to a head:
- The Node 804 was completely out of room. Every bay full, every SSD mount occupied and the array had sprawled to 21 data drives plus dual parity between the main box and the DAS.
- The E3-1270 v2 and 32GB of DDR3 were increasingly the ceiling on everything I wanted to run alongside the array.
- I’ve got a 25U rack sitting in a box waiting to be assembled and a microATX cube case with no rails was never going to be part of that future.
So the mission: move Snuts to a 4U hot-swap chassis on a platform with actual expansion headroom, without touching the DAS or the OTiS box that lives inside it.
For the lineage nerds: I went back and forth on what guide family this build actually belongs to. On build day I was loosely calling it an NK5 spinoff but the honest answer after comparing it against the guides is that this is SNAFU philosophy ported one generation forward. Same used-enterprise RDIMM strategy, same 4U hot-swap chassis approach, same narrow ILM cooler gotcha, just on X10/LGA2011-3/DDR4 instead of X9/LGA2011/DDR3. NK5’s contribution here is the chassis recommendation and that’s about it.
The parts
| Part | Model | Condition | Notes |
|---|---|---|---|
| Chassis | Rosewill RSV-L4412U 4U 12 bay hot-swap | New | |
| Motherboard | Supermicro X10SRi-F rev 1.01B | Used | BIOS 3.4, latest available |
| CPU | Intel Xeon E5-2690 v4 (SR2N2) | Used | 14c/28t, 2.6GHz base, 3.5 turbo, 135W |
| RAM | 4x Micron 32GB DDR4-2400 ECC RDIMM (128GB) | Used | 1DPC quad channel |
| CPU Cooler | Noctua NH-U9DX i4 | New | Narrow ILM brackets, more on this below |
| PSU | EVGA SuperNOVA GT 1000W | New | ECO mode off |
| Front fans | 3x Arctic P12 PWM PST CO | New | Replacing stock 120mm |
| Rear fans | 2x Arctic P8 PWM PST | New | Replacing stock 80mm |
| HBA (bays 1-8) | LSI 9211-8i | Migrated | From the Node 804 |
| HBA (bays 9-12) | LSI 9207-4i4e | New | Bought to bridge the gap, see below |
| HBA (DAS) | LSI 9207-8e | Migrated | Now the sole DAS uplink on a wide port |
| 10GbE | Mellanox MNPA19-XTR | Migrated | Point to point to my PC |
| NVMe cache | 2x Oracle P4608 6.4TB on PLX PEX8718 riser | Migrated | btrfs RAID1 |
| Drive sleds | Zack Freedman printed sleds (Printables 1632299) | Printed | Polymaker ASA on a Bambu P1S |
| IO shield | Printed (Printables 726369) | Printed | Used board didn’t come with one |
(A quick note on the architecture, since this matters for how everything downstream reads: the media library itself spans the whole array, NAS and DAS both, since the DAS drives are live members of the Snuts array. Plex-wise there are two instances. A Docker container on Snuts runs a dedicated Plex server that only ever direct-streams 4K/UHD content, no transcoding, full stop. That 4K library gets added into my main Plex instance on OTiS as a pinned server, visible only on my own logged-in owner account. My actual Plex users log into the OTiS instance and never see the 4K server exists. OTiS handles every transcode and is the only thing any of them ever talk to. The DAS drives being live array members means shutdown order is sacred: Snuts fully off before OTiS ever loses power.)
Enter the L4412U
Fan eviction first, everything else second.
- Tip #1 I learned: The L4412U backplane has fan headers on it but they are power only, no PWM signal. Do not plug your front fans into them unless you enjoy 100% duty cycle forever. Everything went to motherboard headers instead so the BMC can actually control zones (front P12s on FAN2-4 in Zone 0, rear P8s daisy-chained via their built-in PST onto FAN5 in Zone 1).
Board prep
Used board off eBay, which meant two small surprises. One: no IO shield in the box.
Two: the CR2032 was reading 2.907V on the sensor log. Had a fresh one sitting in the parts bin already, so that got swapped before anything else went any further.
- Tip #2 I learned: Supermicro boards of this era use narrow ILM and Noctua’s DX line ships with multiple bracket sets. The NM-XFB5 narrow brackets are the ones you want for the X10SRi-F. I also ruled out the bigger U12DX up front on clearance math alone rather than finding out the hard way inside the case: the U9DX comes in at 125mm and that’s the number I had room for in this 4U.
4x 32GB Micron DDR4-2400 RDIMMs in the outermost slots per the manual, quad channel, one DIMM per channel. Confirmed 128GB at full 2400 on first POST, no surprises.
Into the chassis
I mapped the standoffs before the board went anywhere near the tray. Nine standoffs repositioned for the footprint, nine screws, done. The PSU bay only lines up for 4 of its 6 screws with this EVGA which bothered me for about a minute and then never again.
Backplane power runs off 3 of EVGA’s 6-pin-to-triple-Molex cables, one per backplane module, spread across three independent PSU rails. Used the PERIF port for one and the two SATA ports for the other two. Left the VGA/PCIe modular ports alone since those use a different pinout than what these cables expect.
One card in the stack is new rather than migrated: the Node 804 topped out at 8 bays and one HBA covered all of them. The 4412u’s extra 4 bays needed their own internal SFF-8087 connectivity, so a 9207-4i4e came in specifically to bridge that gap.
- Tip #3 I learned: The rear slot numbering on the L4412U is not what you’d assume. Slot 6 is at the top nearest the IO shield, Slot 1 at the bottom. Worth checking before you plan your card layout since it’s easy to get backwards.
First POST was clean. 128GB at 2400 confirmed, iKVM up immediately, no beep codes, no errors.
The sleds
The stock sleds work but replacements are hard to source and the fit is loose. So in keeping with my apparent lifestyle of printing server parts now, all twelve bays got Zack Freedman’s printed sleds in Polymaker ASA. White bodies for data, teal for parity, so I can tell at a glance which drives will end my whole weekend if I pull the wrong one.
ASA over PETG/PCTG because these live in an enclosed case with warm drives 24/7 and I wanted the heat deflection headroom. Tighter fit than stock means a more positive backplane connection too. Tradeoff is it transfers a bit more vibration to the drives. Accepted.
Migration day
Drive map documented serial by serial before anything moved (21 data + 2 parity across the main chassis and DAS, you do not wing this part), Node 804 shut down clean, USB key and cards over, array up.
Notes on the card stack
Rear slot cards: 9211-8i covering bays 1-8, 9207-4i4e for bays 9-12, 9207-8e as the DAS uplink, Mellanox 10GbE, and the PLX riser carrying both Oracle P4608 NVMe cache drives.
One change worth a mention since it’s tied to this migration: the DAS is now on a proper 8-lane wide port off the 9207-8e instead of the single 4-lane link it was running before, which was bottlenecking parity checks. That’s really a DAS story more than a Snuts-chassis story though, so I’ll leave the full writeup for its own thread if there’s interest.
With the 4i4e’s internal ports live, Slot 9 got a 20TB Exos, cleared natively in-array, formatted and mounted at ~19.6TB usable. Slots 10-12 are next up: fill with 20TB, then start rolling the smallest 14s out of the array.
Final specs
Snuts (this build): RSV-L4412U | X10SRi-F | E5-2690 v4 14c/28t | 128GB DDR4-2400 ECC RDIMM | EVGA GT 1000W | NH-U9DX i4 | 9211-8i + 9207-4i4e (chassis) + 9207-8e (DAS wide port) | 2x Oracle P4608 6.4TB NVMe cache btrfs RAID1 | Mellanox 10GbE | Unraid Pro 7.3.1, 22 data drives + dual parity
OTiS + DAS: unchanged from the previous thread, still running the main client-facing Plex instance, handling every transcode, and quietly pinning in the Snuts 4K server for owner-eyes-only.
Next up is finally assembling the 25U rack that’s been staring at me from its box and getting this thing on rails. The Node 804 gets a teardown and a new home once I’m confident everything’s settled.
Thanks to the guides and the folks here and on Discord. My wallet still hasn’t forgiven any of you.