HPE ProLiant Compute XD685 (HGX H200)
Overview
Performance Metrics
| Precision | Peak (PFLOPS) | Per GPU (TFLOPS) | TFLOPS/W |
|---|---|---|---|
| FP64 | 0.3 | 34.0 | — |
| FP32 | 0.5 | 67.0 | — |
| TF32 | 4.0 | 494.5 | — |
| FP16 | 7.9 | 989.5 | — |
| BF16 | 7.9 | 989.5 | — |
| FP8 | 15.8 | 1979.0 | — |
| INT8 | 15.8 | 1979.0 | — |
All figures are dense. Vendors commonly headline the sparse number, which is twice the dense one.
Configuration
- Accelerator
- H200 SXM 141GB
- Count
- 8
- Memory per GPU
- 141 GB
- Bandwidth per GPU
- 4.8 TB/s
Power & Efficiency
- Aggregate Power
- —
- TDP per GPU
- 700 W
- FP8 Efficiency
- —
- Aggregate Bandwidth
- 38.4 TB/s
System Details
Interconnect
NVLink 4.0 / NVSwitch 3.0 across the HGX H200 baseboard; 4x FHHL PCIe Gen5 x16 slots for fabric, NIC or DPU
Notes
The eight-way XD685 in its NVIDIA HGX H200 configuration, the only one of the four HPE offers in both air-cooled and liquid-cooled form: 6U for air with accelerator kit S3P89A, 5U for direct liquid cooling with S4X53A. Two fifth-generation AMD EPYC processors drive it, from 32 to 96 cores and 280 to 400 W, across 24 DIMM slots with twelve channels per socket and up to 6 TB of DDR5-6400. Storage is eight EDSFF NVMe E3.S drives, or twelve with a storage controller, plus RAID-1 M.2 boot. No chassis part number is recorded because two apply depending on cooling, S4Q28A for air and S4R96A for liquid. HPE publishes no system power draw for the XD685, only twelve 3,000 W Titanium supplies with ten deployable, so nothing is summed here; the compute figures are NVIDIA's for the HGX H200 baseboard rather than an HPE claim.