Building February

The hardware has been sitting in boxes for weeks. Yesterday was the day February was going to become a server rather than a collection of parts on the floor. I cleared the workspace, laid everything out, and started building.

This is the honest account of how that went.

The hardware

The Fractal Define 7 XL is a serious case. Eight dedicated HDD trays, room for multiple SSDs, excellent cable management channels, and enough physical space to work in comfortably. It is the right case for a build like this. Working inside it is a pleasure compared to the cramped interiors of smaller enclosures. That much went well.

The build

CPU, cooler, and RAM went onto the motherboard before it went into the case. Mounting a cooler in a case is significantly more awkward than mounting it on a bench. The Ryzen 7 5700X seated cleanly, the Corsair sticks clicked into place, and the board went into the Define 7 XL without drama.

The NVMe drives went into their M.2 slots on the motherboard. The two Crucial SSDs were mounted in the dedicated SSD bays in the case. All eight Ironwolf HDDs went into the eight dedicated trays, which the Define 7 XL handles elegantly: tool-free mounting, clean alignment, no hunting for screws.

The two Axagon PCES-SA4X4 HBA cards went into the PCIe slots. This is where the first constraint emerged.

The PCIe problem

The ASUS ROG Strix B550-F has three PCIe slots: one x16 (PCIe 4.0), one physical x16 slot wired as x4 (PCIe 3.0), and one x1. The Axagon PCES-SA4X4 requires a minimum of an x4 slot and uses x2 lanes. Two HBA cards means two slots of x4 or larger. That accounts for both the x16 and the x4 slot.

The GTX 1080 has nowhere to go. The only remaining slot is the x1, which cannot physically accept a full-size GPU. This was a planning oversight. February is intended to eventually run GPU-accelerated workloads, but for now it will run headless. The GPU stays on the shelf until a different solution is found, whether that is a different motherboard further down the line or accepting that GPU passthrough is not part of this particular build.

It is a frustration, but not a blocker for the immediate goal: a Proxmox server with a large ZFS storage pool.

The GTX 1080 has nowhere to go. This was a planning oversight that will need a different solution.

First power-on: partial success

With everything installed except the HDD SATA data cables, the system powered on. POST completed. BIOS detected the CPU, all 128GB of RAM, both NVMe drives, and both HBA cards. That was genuinely satisfying after a full day of building.

Then I connected the SATA data cables and power to all eight Ironwolf HDDs, the two Crucial SSDs, and attempted to boot again.

Nothing. The system would not power on.

The power problem

Methodical elimination pointed at the drives. Disconnecting all of them restored the ability to power on. Adding them back one category at a time narrowed it down: the eight Ironwolf HDDs, when all connected simultaneously, prevented the system from starting.

The likely cause is spin-up surge. Eight Seagate Ironwolf drives spinning up simultaneously can draw 160 to 240 watts in the first two to three seconds of startup. The PSU’s overcurrent protection on the 12V rail is seeing that surge and shutting down before anything else can happen.

There is a second contributing factor specific to the Axagon PCES-SA4X4 cards. These HBAs have no auxiliary power connector. They draw everything through the PCIe slot, which is rated for 75 watts. Four Ironwolf drives spinning up through a single 75-watt PCIe slot is asking too much of that connection, regardless of what the PSU as a whole can supply.

The 1200-watt PSU is not the problem. The problem is how the power reaches the drives and what happens in the first few seconds of startup.

Where things stand

February is partially built and partially working. The system boots cleanly without the HDDs connected. The NVMe drives are detected. The HBA cards are recognised. The RAM and CPU are correct.

The eight Ironwolf HDDs are the outstanding problem. The paths forward are:

Staggered spin-up, if it can be enabled either in the BIOS or via the HBA card settings, would spread the startup current draw over several seconds rather than concentrating it in one simultaneous surge. This is the first thing to try, and it costs nothing to attempt.

Connecting fewer drives per HBA card at startup, then adding the remainder once the system is running, is a viable workaround if staggered spin-up is not available. It is not elegant but it is functional.

Replacing the HBA cards with models that include auxiliary power connectors would solve the per-slot power problem properly. The Axagon cards were chosen for their price and simplicity, but they are not the right choice for eight spinning drives. This is the longer-term solution if the staggered spin-up approach does not work reliably.

February exists. It is not finished. That is an honest place to be.

What I learned

Plan PCIe slot allocation before buying the GPU. The B550-F’s slot configuration is not unusual but I did not map it against the HBA requirements until the cards were in the machine. Doing that on paper first would have caught the conflict.

Spin-up surge is a real thing and eight drives is a lot of drives. The spec sheet total wattage is not the number that matters at startup. The surge current in the first two seconds is what trips protection circuits, and that number is not always listed prominently in the documentation.

The Fractal Define 7 XL is an excellent case for a build like this. Everything about working inside it was straightforward. Cable routing is clean, the HDD trays are well-designed, and there is enough space to actually see what you are doing. If you are building a storage-heavy server, this case earns its size.

February will get there. The next build session will focus on staggered spin-up and getting all eight drives recognised. When that works, Proxmox goes on.