Supermicro sells boards, not sealed systems — and that difference defines its memory story. Where Dell and HPE buyers inherit a configured machine, Supermicro builders choose every component, which makes 'what memory does this board take' a first-class purchasing decision rather than an upgrade afterthought. This hub covers the boards our tracker can serve properly: verified specs from Supermicro's own product pages, wired to live cross-brand DDR4 pricing.
The boards below are the homelab and small-business canon: the X11 single-socket Xeon Scalable pair (X11SPL-F and X11SPi-TF) that anchors countless storage servers and Proxmox hosts, the X12SPi-TF that brings the formula to Ice Lake and DDR4-3200, and the H12SSL-i — the community's favorite door into single-socket AMD EPYC. All four run DDR4 ECC, which in 2026 means all four eat from the affordable side of the memory shortage.
Supermicro's board pages state memory support plainly — slot count, capacity ceiling, module types, speed grades — and our rules boxes carry exactly that, with one honest wrinkle disclosed per board: where a page enumerates a speed range (the X11 boards list 2133 through 2933), our tables accept that range; where it states a single grade (the X12 and H12 pages say DDR4-3200, full stop), our tables hold to that grade rather than assuming slower modules are blessed. Conservative, but traceable to the page we cite.
The firmware footnotes matter more on Supermicro than elsewhere because boards live multiple lives: the X11 pair requires BIOS 3.0a or later for 2nd-gen Xeon Scalable (and with it 2933 MT/s operation), and the H12SSL-i wants BIOS 2.3+ before it will host EPYC 7003 parts with 3D V-Cache. A used board is a used firmware state — check before buying the fast silicon or the fast memory.
Channel architecture is the performance story here. The Xeon Scalable boards run six channels per socket; the EPYC H12 runs eight, and cares about them intensely — EPYC's bandwidth collapses on partially populated channels, which is why the H12SSL-i page's scenarios default to eight matched modules and why 8× 16GB genuinely outperforms 4× 32GB on that platform despite identical totals. The per-board pages carry the specifics; the principle is universal: feed every channel the board gives you.
Sourcing for these boards is the pure cross-brand market — there is no vendor-branded DIMM program to pay for, just Samsung, SK Hynix, Micron and Kingston parts at whatever the market charges today. That makes the live $/GB tables on the board pages the entire purchasing decision, and it makes the used-pull market — fed by the same Xeon Scalable decommissioning wave that supplies the CPUs — the natural first stop for X11-class builds.
Using this hub: builders comparing boards should scan the list below for compatible-module counts and best live $/GB — a board whose memory market is deep stays cheap to grow into. Owners upgrading in place should go straight to their board's page: verified rules on top, live table underneath, upgrade scenarios computed from real listings at the bottom. All of it inherits the site's rules — 1024-based $/GB, out-of-stock excluded, freshness signals bound to the real refresh, and nothing stated that the board page doesn't state.
The board-buyer's memory sequence differs from the system-buyer's in one important way: nothing arrives pre-validated. That makes the verification order matter — board page first (type, grade, capacity ceiling), BIOS revision second (generation support hides there), memory third. The model pages here carry the first two so the third is a price decision, which is how a component-market purchase should feel.
Supermicro's speed-listing style splits our coverage into two honest modes, worth understanding as a buyer: boards whose pages enumerate a range (the X11 pair lists 2133 through 2933) get tables spanning that range, while boards stating a single grade (X12SPi-TF and H12SSL-i say 3200, full stop) get tables holding that line. Where the page is terse, we are strict — the alternative is inventing support the manufacturer never claimed, which is how compatibility folklore starts.
The Intel-versus-AMD fork inside this hub is really a channels-versus-simplicity fork: the Xeon Scalable boards tolerate incremental population gracefully, while the EPYC H12's eight channels make full population the performance baseline. Budget accordingly — an H12 build's memory line is 8× whatever module wins the table, and pretending otherwise just defers the second order.
IPMI deserves a memory-specific mention because it changes the used-module risk calculus on every board here: Supermicro's BMC exposes per-slot ECC counters out of the box, which means a tested pull is not a leap of faith but a monitored component whose first misstep you will see in a log before you feel it in a workload. Set the alert, buy the pull, save the difference.
Last, the meta-point this vendor teaches: because Supermicro publishes real specs on public product pages, its platforms are the easiest in this cluster to verify and the hardest to be misled about. The rules boxes here quote those pages directly, link them beneath every table, and add nothing — which is the standard the whole server-memory market should be held to.
On warranties: Supermicro boards carry their own warranty independent of the memory installed, and the tested-memory lists are guidance rather than gatekeeping — standard modules of the documented spec neither void coverage nor require special blessing. The practical protection is the one already described: IPMI's ECC counters, watched from day one, catch the rare bad module while it is still a return rather than an outage.
Board revisions add one caution the used market teaches: Supermicro boards iterate quietly (Rev 1.01 versus 2.0 can differ in supported CPU steppings and occasionally memory behavior), and the product page describes the current revision. When buying a used board, confirm the revision printed on the PCB against the page's notes before assuming every footnote applies — five seconds of reading that has saved many builders a return cycle.
The scenarios on each board page stay inside the board's slot count and price only in-stock listings — for the H12 that means eight-module baselines, for the Xeon boards the incremental rungs builders actually take. As everywhere on this site, they are computed baselines, not quotes: the tables refresh every few hours, and the patient builder who watches a specific part number's chart wins against every scenario we print.
And for the reader who arrived here mid-build with a board already on the bench: go straight to its page, read the rules box against the sticker on your PCB, and buy from the table with confidence — that thirty-second check is what this hub was built to make possible.
They run standard DDR4 ECC modules of the documented type and grade; Supermicro publishes tested-memory lists per board for those who want validated part numbers. In practice, matching type, speed and capacity from the board page's memory section — which is what our tables enforce — is what compatibility requires.
For Intel builds, the X11SPL-F remains the price-performance anchor: cheap used Xeon Scalable silicon, eight DDR4 slots, IPMI, and the market's cheapest registered ECC. For PCIe lanes and core density, the H12SSL-i's single-socket EPYC is the step up — at the cost of needing eight matched DIMMs to perform.
EPYC 7002/7003 runs eight memory channels per socket, and every empty channel forfeits an eighth of the bandwidth. The board functions with fewer modules, but memory-bound workloads feel it directly — which is why pricing 8× smaller modules against 4× larger ones (the live table makes this easy) usually favors the eight-way split.
The product pages for both boards state DDR4-3200 without enumerating lower grades, so our tables list only 3200-rated modules — we don't assume support the page doesn't claim. The X11 boards are the opposite case: their pages list 2133 through 2933, and our tables accept that full range.
Yes, with the standard discipline — DRAM doesn't wear, so tested Xeon-Scalable-era pulls are the natural fuel for X11 builds especially. Run memtest86+ before trusting a lot, watch ECC counters early, and buy matched sets per channel. The savings against new stock routinely fund the rest of the build.