NVIDIADatacenter

NVIDIA DGX H100

Eight H100s, one NVLink domain, zero integration headaches.

$300,000

8x H100 SXM5 (640GB HBM3 total)

GPUs

2TB DDR5

System Memory

~10.2 kW

Power Draw

Fully integrated 8x H100 SXM5 system with NVSwitch fabric, dual Xeon Platinum hosts, and 400Gb/s-per-GPU networking headroom. The reference design most DGX PODs are built from.

Full Specifications

Compute

GPUs: 8x H100 SXM5 (640GB total)

Memory

System Memory: 2TB DDR5

Interconnect & Networking

Fabric: 4x NVSwitch, 900 GB/s per GPU
Networking: 8x 400Gb/s OSFP (ConnectX-7)

Power & Cooling

Power Draw: ~10.2 kW

General

CPU: 2x Intel Xeon Platinum 8480C

Compatible & Required

Required — won't function without

3-Phase PDU$9,500

Pairs well with

Quantum-2 Switch$52,000
Flash Storage Node$190,000
ConnectX-7$2,500

Add software & support

DGX OS
Free
NVIDIA AI Enterprise
$4,500/yr
1-Year Standard Support
$2,500
3-Year Priority Support (24/7 SLA)
$12,000