Intel Xeon Phi 7295 socket: LGA 3647
The Intel Xeon Phi 7295 uses the LGA 3647 socket on HPC server boards. It is a 72 core Knights Mill flagship at 320 W with boost to 1.6 GHz.
- Socket
- LGA 3647
- Launched
- December 2017
- Cores and threads
- 72 and 288
- Base power
- 320 W
- Codename
- Knights Mill
- Segment
- Server
- Family
- Intel Xeon Phi
- Clocks
- 1.5 / 1.6 GHz
Intel Xeon Phi 7295 specifications
| Cores | 72 |
|---|---|
| Threads | 288 |
| Core architecture | Knights Mill on 14 nm |
| Clocks | 1.5 GHz base, up to 1.6 GHz Turbo |
| Cache | 36 MB L2 |
| On-package memory | 16 GB MCDRAM, 400+ GB/s |
| Processor base power (TDP) | 320 W |
| Manufacturing process | 14 nm |
| Memory support | DDR4-2400, six channels, up to 384 GB with ECC |
| PCI Express | PCIe 3.0, 16 lanes from the CPU |
| Integrated graphics | None |
| Multiplier | Locked |
| ECC memory | Yes |
| Package | FCLGA3647 (76.0 × 56.5 mm) |
Motherboard compatibility
Intel Xeon Phi 7295 needs a motherboard with a LGA 3647 socket. Chipset by chipset data for this codename is not recorded yet. See the LGA 3647 chipset list and the board maker's CPU support list.
LGA 3647 at a glance
| Type | LGA, 3,647 contacts |
|---|---|
| Package size | 76.0 × 56.5 mm |
| Introduced | 2016 |
| Memory | DDR4-2933 ECC, six channels, up to 2 TB |
| Cooler mounting holes | Board specific (narrow and square ILM heatsinks) |
What the Xeon Phi 7295 is
The Xeon Phi 7295 is a seventy two core, two hundred eighty eight thread flagship accelerator from Knights Mill. It launched in December 2017 with a 1.5 GHz base and up to 1.6 GHz Turbo at 320 W. Seventy two cores cover top training nodes where maximum throughput matters.
Thirty six megabytes of shared L2 cache feed seventy two cores. The 1.5 GHz base leads Phi parts on clocks, while 1.6 GHz burst helps single thread setup. Three hundred twenty watts demand the highest airflow in the Phi set.
Sixteen gigabytes of MCDRAM plus six channel DDR4-2400 up to 384 GB hold large models. MCDRAM flat mode gives explicit placement for tuned training loops. Cache mode speeds untuned code.
Knights Mill vector neural instructions accelerate mixed precision integer math for training. Throughput per watt suits data center training before later accelerators arrived. Locked multiplier keeps large clusters uniform.
Socket and motherboard compatibility
The 7295 fits LGA 3647 on HPC boards with flagship power delivery for 320 W. BIOS must list 72 core Phi parts. Single socket nodes are the design.
MCDRAM modes follow Intel tuning guides for training frameworks. DDR4-2400 ECC covers dataset capacity. The Socket vs chipset: what decides whether a CPU works guide covers server socket use.
The Xeon Phi 7285 on the same socket trims cores and power where cost matters. There is no higher Knights Mill above the 7295.
Memory, PCIe and cooling
DDR4-2400 ECC in six channels up to 384 GB plus 16 GB MCDRAM is the flagship design. Bandwidth peaks near 115.2 GB per sec on DDR4 plus 400+ GB/s on MCDRAM. ECC protects week long runs.
Sixteen PCIe lanes wire fabric and storage. Training fabrics take x16 slots. Node local NVMe holds datasets.
Three hundred twenty watts need liquid or top air with high static pressure. Cool ambients sustain boost best.
Upgrade paths
On LGA 3647 there is no higher Phi above the 7295, so growth means more nodes or newer accelerators on other platforms.
Fabric bandwidth and storage upgrades often help training clusters more than CPU swaps.
Frequently asked questions
What socket does the Xeon Phi 7295 use?
It uses LGA 3647 on HPC server boards.
How many cores does the Phi 7295 have?
Seventy two cores with two hundred eighty eight threads at 1.5 GHz base with up to 1.6 GHz Turbo.
What is MCDRAM on the Phi 7295?
16 GB of on package memory at 400+ GB/s, plus DDR4-2400 up to 384 GB.
When did the Phi 7295 launch?
December 2017 as the Knights Mill flagship.
Sources and verification
Facts last checked against these sources on . Page last edited on .