2.3 GHz / 16 Cores / 32 Threads / 22 MB Cache / 125 W / 4-Socket
| Socket | LGA3647 |
|---|---|
| Series | Xeon Gold |
| Generation | 2nd Gen (Cascade Lake) |
| Cores | 16 |
| Memory Type | DDR4 RDIMM · DDR4 LRDIMM · DDR4 3DS RDIMM |
| Max Memory Speed | 2667 MHz |
| PCIe Version | PCIe 3.0 |
| Lithography | 14 nm |
| Memory Channels | 6 channels |
| Condition | New · Used |
| Threads | 32 |
| Base Frequency | 2.3 GHz |
| Max Turbo Frequency | 3.9 GHz |
| Cache | 22 MB |
| TDP | 125 W |
Intel Xeon Gold 5218 — 16 cores and 32 threads at 2.3 GHz base / 3.9 GHz turbo, 22 MB cache, 125 W TDP, six-channel DDR4-2667 and four-socket scalability in FCLGA3647 packaging. Supplied new or as tested pulls, with a 1-year warranty.
The 5218 is the sixteen-core Gold that most two-socket general-purpose servers were actually specified with, and the reason is arithmetic rather than marketing. It carries 1.375 MB of cache per core — 22 MB over sixteen — which is exactly the per-core figure of the eighteen-core 5220, 5220S and 5220T. Same 3.9 GHz turbo ceiling, same six channels at 2667, same 125 W and 87 °C, same single AVX-512 FMA unit. It is the same recipe with two fewer cores.
What that buys is base clock and price. At 2.3 GHz it runs 100 MHz faster on the base clock than the eighteen-core 5220, which is not the direction most people expect: fewer cores in the same 125 W envelope means more frequency per core. Per-socket all-core throughput is still lower than the 5220, but single-thread and lightly-threaded work is slightly quicker, and the price tier is below the eighteen-core parts.
125 W against an 87 °C case limit is 0.42 °C/W of heatsink allowance from a 35 °C inlet — the same figure as the 5220 and 5220S, which means any chassis or heatsink already validated for a standard 125 W Gold takes this processor without a cooling review. The 150 W 6240 wants 0.333 and the 200 W 6254 wants 0.235; this part sits well inside both.
Cache is 22 MB against 24.75 MB on the eighteen-core parts, and memory is the same DDR4-2667 across six channels — 128 GB/s theoretical to 1 TB per socket, with Optane persistent memory supported. If the workload is bandwidth-bound, note that the 6000-series parts run 2933 and give about 9 % more; if it is not, the 2667 ceiling costs nothing.
AVX-512 is present with one FMA unit per core, the Gold 5000-series rule (5122 and 5222 are the documented exceptions with two). For virtualisation, web and application tiers, databases, file serving and container platforms that is irrelevant. For dense linear algebra, HPC and inference kernels compiled around 512-bit vectors it is decisive, and a Xeon Gold 6240 with its second FMA unit and 2933 memory is the correct purchase instead.
| Product collection | 2nd Gen Intel Xeon Scalable (Cascade Lake) |
|---|---|
| Processor number / sSpec | 5218 / SRF8T (B1 stepping; the character after SRF is a digit eight, not a letter B) |
| Ordering code | CD8069504193301 (tray) / BX806955218 (boxed) |
| Cores / Threads | 16 / 32 (Hyper-Threading) |
| Base / Max turbo frequency | 2.3 GHz / 3.9 GHz |
| Cache | 22 MB (1.375 MB per core) |
| TDP | 125 W (TCASE 87 °C) |
| Socket | FCLGA3647 (FC-LGA14B, 76.0 mm × 56.5 mm) |
| Memory | 6 channels DDR4-2667, up to 1 TB per socket, ECC RDIMM / LRDIMM / 3DS RDIMM, Optane persistent memory supported |
| Scalability | 4S (up to four sockets), 2 UPI links |
| PCIe | 3.0, 48 lanes |
| AVX-512 | One FMA unit per core (Gold 5000-series rule; 5122 and 5222 are the exceptions with two) |
| Speed Select | SST – Performance Profile: no; SST – Base Frequency: no |
| Lithography | 14 nm |
| Launch | Q2 2019 |
| Status | Discontinued (servicing updates issued through 30 June 2025) |
| Embedded options | Not available |
| Accelerators | AVX-512 (1 FMA), Intel DL Boost (VNNI), Intel RDT, Intel VMD, Run Sure |
Any FCLGA3647 board whose BIOS carries Cascade Lake microcode, in one-, two- and four-socket form. The configurations this part is normally bought for:
Before ordering, confirm the BIOS revision — boards that shipped before the 2019 launch usually need an update before they will POST. Confirm the memory population too: the part runs at 2667, so 2933 modules fit but deliver no more bandwidth than the processor limit. And if the platform is a Gen10 Plus or a 15th generation PowerEdge, it uses LGA4189 and this processor will not fit.
| Processor | Generation | Cores / Threads | Base / Max turbo | Cache | TDP | Memory support |
|---|---|---|---|---|---|---|
| Xeon Gold 5218 | 2nd Gen (Cascade Lake) | 16 / 32 | 2.3 / 3.9 GHz | 22 MB | 125 W | 6ch DDR4-2667 · 1 TB · 4S |
| Xeon Gold 5220 | 2nd Gen (Cascade Lake) | 18 / 36 | 2.2 / 3.9 GHz | 24.75 MB | 125 W | 6ch DDR4-2667 · 1 TB · 4S |
| Xeon Gold 5220S | 2nd Gen (Cascade Lake) | 18 / 36 | 2.7 / 3.9 GHz | 24.75 MB | 125 W | 6ch DDR4-2667 · 1 TB · 4S |
| Xeon Gold 5220T | 2nd Gen (Cascade Lake) | 18 / 36 | 1.9 / 3.9 GHz | 24.75 MB | 105 W | 6ch DDR4-2667 · 1 TB · 4S |
| Xeon Gold 6230 | 2nd Gen (Cascade Lake) | 20 / 40 | 2.1 / 3.9 GHz | 27.5 MB | 125 W | 6ch DDR4-2933 · 1 TB · 4S |
| Xeon Gold 6240 | 2nd Gen (Cascade Lake) | 18 / 36 | 2.6 / 3.9 GHz | 24.75 MB | 150 W | 6ch DDR4-2933 · 1 TB · 4S |
| Xeon Gold 6252 | 2nd Gen (Cascade Lake) | 24 / 48 | 2.1 / 3.7 GHz | 35.75 MB | 150 W | 6ch DDR4-2933 · 1 TB · 4S |
Against the eighteen-core parts the 5218 trades two cores and 2.75 MB of cache for 100 MHz more base clock and a lower price tier, at identical wattage, case limit, memory speed and turbo ceiling. Per-core throughput is therefore slightly better here, per-socket throughput slightly worse — roughly 11 % fewer threads at about the same clock. Thread-dense estates should take the 5220; single-thread-sensitive ones do not lose anything.
Against the 5220S the gap is 400 MHz of base clock on the same 125 W and 87 °C, plus two cores and 2.75 MB. The S is the fast eighteen-core bin and is priced accordingly.
Against the 6230 you give up four cores, 5.5 MB, 2933 memory and 100 MHz of turbo, for 200 MHz of base clock at the same 125 W — and typically a lower price. It is the clearest fork in this range: threads against clock, both thermally identical, and for container or virtualisation density the 6230 usually wins.
Against the 6240 two things change at once: 25 W more heat and the second AVX-512 FMA unit, plus 2933 memory and 300 MHz more base clock. If the workload touches 512-bit vectors, that second FMA is not a marginal gain — it is a different class of machine. Against the 6252 the step is to 24 cores and 35.75 MB at 150 W, which needs the higher-performance heatsink; the 6254 goes further again to 3.1 GHz base and 200 W.
Every processor carries a 1-year warranty against defects in workmanship and function. Verified faulty units are replaced or refunded and we cover return shipping. Stock is held in both new and used condition; the condition you order is confirmed before dispatch, and used units are pull-tested stock rather than untested lots. Stock is held in our own warehouse and normally dispatches within three to seven working days, with tracking issued at dispatch.
Pricing is quoted per order because condition and quantity both move it. Send the quantity, the condition you need and the server model the part is going into, and you will get a quote with lead time the same working day. Matched pairs for two-socket builds are available on request — tell us if the processors are going into the same chassis and we will supply them from the same stepping.
Tell us who you are — we reply within one working day with a firm quote, stock status and lead time. Volume discounts available.
2.1 GHz / 8 Cores / 16 Threads / 11 MB Cache / 85 W / 78 °C TCASE
2.2 GHz / 8 Cores / 16 Threads / 11 MB Cache / 70 W / 91 °C TCASE
2.3 GHz / 10 Cores / 20 Threads / 13.75 MB Cache / 95 W / 91 °C TCASE
2.4 GHz / 10 Cores / 20 Threads / 13.75 MB Cache / 100 W / 2-Socket