NVIDIA HGX platform server
NVIDIA HGX B200
8 × NVIDIA Blackwell SXM · 8U rack server
Baseline configuration
This is the reviewed baseline for discussion. Alternative components are not listed because they have not been validated; any change is confirmed with the quote.
Accelerators
- Primary GPUs
- 8 × NVIDIA Blackwell SXM
- GPU memory
- 180 GB per primary GPU · HBM3e
- GPUs included
- Included
System
- Platform
- NVIDIA HGX platform server
- Form factor
- 8U rack server
- CPU
- 2 × AMD EPYC 9555 · 64 cores each
- System memory
- 24 × 96 GB DDR5-6400 ECC · 2,304 GB
- Storage
- 1 × 2 TB M.2 PCIe 4.0; 2 × 7.68 TB U.2/U.3 PCIe 5.0
Facility
- Networking
- Dual 1GbE plus management; high-speed networking requires confirmation
- Power
- 12 × 3,000 W Titanium PSUs · 6+6 redundancy · estimated draw 14.3 kW
Notes
- AI Enterprise is not included.
For discussion only. Pricing, GPU inclusion, final specifications, availability, shipping, taxes and procurement eligibility must be confirmed. Selecting a machine makes no commitment. Memory across multiple GPUs is aggregate capacity, not a single shared memory pool.