Which Supermicro family? Hyper vs Ultra vs CloudDC vs Twin vs GPU vs A+ vs SuperStorage

Which Supermicro family should you buy?

Supermicro sells one chassis-and-board catalogue sliced into workload-optimised families. The family letter in the part number is the buying decision. Here is how they line up.

The one-screen selector

Family	Sockets / form	Best for	Trade-off
Hyper (H)	Dual, 1U/2U	Max memory + I/O headroom, GPUs, demanding mixed enterprise	Highest price of the general-purpose lines
Ultra (U)	Dual, 1U/2U	The X11/X12 refurb workhorse — proven, plentiful	Newest generation is X12 (DDR4); not carried to X13
CloudDC (C)	Single/dual, 1U/2U	Value cloud & data-centre nodes, OCP networking	Fewer PCIe/DIMM slots than Hyper
GrandTwin (GT)	4× single-socket in 2U	Per-core-licensed, dedicated hosting, edge	Single socket per node caps per-node RAM/cores
BigTwin (BT)	4× dual-socket in 2U	Max compute density, HCI, scale-out VM farms	Shared power/cooling; serviced as an enclosure
GPU (G/GE)	Dual + 4U/8U	AI training/inference — PCIe (up to 10) or HGX SXM (8-way)	Power and cooling budget (5–18 kW)
A+ (EPYC)	Single/dual SP5	Highest core count and PCIe-5.0 lane count per socket	AMD ecosystem; fewer refurb units than Intel
SuperStorage (SSG)	Dual, 2U–4U	Capacity — 12 to 90+ bays, SDS, backup, object	Compute is secondary to spindle/flash count

How the general-purpose lines actually differ

Hyper is the successor to Ultra. Ultra was the enterprise flagship through X11 (SYS-2029U) and X12 (SYS-620U). At X13, Supermicro did not ship an Ultra — the dual-socket, maximum-memory, maximum-I/O role passed to Hyper (SYS-221H / SYS-621H). So if someone asks for an "X13 Ultra," the honest answer is: that is now the Hyper line. Ultra remains the right refurb pick precisely because its newest generation is X12 (Ice Lake, DDR4) — a mature, well-supplied, cost-effective platform.

CloudDC is Hyper with the fat trimmed. Same X13 board generation and Socket E, but tuned for cost-per-node at cloud scale: OCP 3.0 AIOM networking standard, a leaner PCIe/DIMM count, and 3.5"-capacity front bays. Choose it when you are deploying many identical nodes and do not need Hyper's slot headroom.

Twins trade serviceability for density. GrandTwin puts four single-socket nodes in 2U; BigTwin puts four dual-socket nodes in 2U. Both share power and cooling, so you plan and service at the enclosure level. GrandTwin suits per-core-licensed and dedicated-hosting fleets; BigTwin suits the densest compute and hyperconverged clusters.

Intel or AMD?

The A+ (EPYC) line mirrors the Intel families one-for-one — there is an EPYC Hyper (AS-2025HS) and an EPYC CloudDC (AS-1115CS). Reach for A+ when the workload is core-bound or I/O-lane-bound: current EPYC (H13, Socket SP5) offers up to 160 cores and 128 PCIe 5.0 lanes per socket, ahead of the equivalent Xeon. Stay on Intel SuperServer where the buyer's software is Xeon-validated, where Optane/AMX features matter, or — in the SA refurb market — where second-hand supply is simply deeper.

The refurb-market read (South Africa)

• X12 Ultra / CloudDC is the current price/performance sweet spot: DDR4-3200, PCIe 4.0, strong dual-socket value.
• X11 Ultra (2029U/6029U) is the budget floor — cheap and everywhere, but DDR4-2933, PCIe 3.0 and no manufacturer warranty.
• X13 Hyper / CloudDC and H13 A+ are early-cycle: available new, with the first refurb units appearing as hyperscalers rotate stock.
• HGX-class GPU (8U, 8-way SXM) rarely appears as genuine refurb and carries an 18 kW power envelope — document it as a spec reference rather than assuming stock.