HPE ProLiant Compute XD685 (HGX B300)
Overview
Performance Metrics
| Precision | Peak (TFLOPS) | Per GPU (TFLOPS) | TFLOPS/W |
|---|---|---|---|
| FP64 | 10 | 1.3 | — |
| FP32 | 600 | 75 | — |
| TF32 | 9,000 | 1,125 | — |
| FP16 | 18,000 | 2,250 | — |
| BF16 | 18,000 | 2,250 | — |
| FP8 | 36,000 | 4,500 | — |
| FP4 | 108,000 | 13,500 | — |
| FP6 | 36,000 | 4,500 | — |
All figures are dense. Vendors commonly headline the sparse number, which is twice the dense one.
Configuration
- Accelerator
- B300 SXM 262GB
- Count
- 8
- Memory per GPU
- 263 GB
- Bandwidth per GPU
- 8.0 TB/s
Power & Efficiency
- Aggregate Power
- —
- TDP per GPU
- 1400 W
- FP8 Efficiency
- —
- Aggregate Bandwidth
- 64.0 TB/s
System Details
Interconnect
NVLink 5.0 / NVSwitch 4.0 across the HGX B300 baseboard; 4x FHHL PCIe Gen5 x16 slots for fabric, NIC or DPU
Notes
The eight-way XD685 in its NVIDIA HGX B300 configuration, which requires the high-power liquid-cooled chassis S6Y07A rather than the standard DLC one. This is the denser sibling of the air-cooled XD690, which reaches the same eight Blackwell Ultra GPUs in 10U where this does it in 5U. Two fifth-generation AMD EPYC processors, 24 DIMM slots and up to 6 TB of DDR5-6400, eight EDSFF NVMe E3.S drives. HPE publishes no system power draw for the XD685, only twelve 3,000 W Titanium supplies with ten deployable, so unlike the XD690 no figure is recorded; the compute shown is NVIDIA's specification for the HGX B300 baseboard rather than an HPE claim.