Supermicro

Why GPU and storage builders choose Supermicro

At Dell, HPE and Lenovo a server is a finished product with a fixed options list. Supermicro thinks in building blocks: the same motherboard goes into several chassis, backplanes and risers can be swapped, and many components are industry standard – Broadcom HBAs, ordinary PCIe cards, drives without vendor-specific firmware. That brings three strengths:

• GPU density: systems such as the SYS-4029GP-TRT take eight double-width, passively cooled GPUs, with power supplies and airflow designed for them. In the dual-root version, half the cards connect to each CPU via PCIe switches; single-root variants hang all cards off one processor, which speeds up direct GPU-to-GPU traffic. Suitable cards are under server GPUs.
• Drive density: storage chassis with 36 bays in 4U, populated front and rear, attached to an HBA in IT mode – a good fit for ZFS, TrueNAS and Ceph, where the OS should see the disks directly.
• Open firmware: Supermicro generally offers BIOS and BMC updates for download without a support contract.

The flip side: documentation is thinner, and with so many variants you check compatibility yourself – which GPU power cables a chassis comes with, say, or which backplane is fitted. Ready-built systems are under GPU servers and storage servers.

IPMI: Supermicro's remote management

Supermicro simply calls its management controller IPMI. It has its own network port, which on many boards can optionally share a LAN port with the host, plus a web interface with sensors, event log and remote console; newer firmware also speaks Redfish. With ipmitool you can power and monitor servers from scripts. The graphical console and virtual media usually work without an extra licence. How that compares with iDRAC and iLO is covered in our guide Remote management: iDRAC, iLO and IPMI.