Hardware / GPU servers / HGX B200

Spec sheet · Blackwell · 2024

HGX B200 SXM6: 8-GPU Server System

First-generation Blackwell: eight B200 GPUs with 180 GB of HBM3e each, NVLink 5 at 1.8 TB/s per GPU, and native FP4.

Memory per GPU180 GB1,440 GB per node
Bandwidth per GPU8 TB/sHBM
FP8 dense per GPU4.5 PFFP4 9.0 PF
Max power14.3 kWper 8-GPU node
HGX B200 SXM6: 8-GPU Server System

Full specifications

Compute

GPUs
8× NVIDIA B200 SXM6
FP4 (system)
~72 PFLOPS dense
FP8 (system)
~36 PFLOPS dense

Memory

GPU memory
180 GB HBM3e per GPU · 1.4 TB total
Memory bandwidth
8.0 TB/s per GPU · 64 TB/s total

Interconnect

GPU interconnect
NVLink 5 · 1.8 TB/s per GPU · NVSwitch
Network fabric
8× ConnectX-7 400G NDR InfiniBand or Ethernet

Power and cooling

GPU TDP
1,000 W per GPU
System power draw
14.3 kW max (DGX B200)

Platform

Host platform
2× Xeon SP or EPYC · 2 TB DDR5 · PCIe Gen 5
Common chassis
Supermicro SYS-821GE-TNHR · Dell XE9780
Form factor
10U air-cooled · 4U liquid-cooled variants
Architecture
Blackwell
First shipped
2024

Power, cooling and rack density

CoolingPer nodePer rackRack power
Air14.3 kW2 nodes29 kW
Liquid14.3 kW6 nodes86 kW

A 32-node cluster (256 GPUs) needs about 0.7 MW of facility power: a colocation cage or suite. Size your own cluster in the site planner →

What one node can serve

Largest open models that fit on eight GPUs at FP8 with an 8K context, with estimated output throughput.

  • DeepSeek-V4.1 Flash12K tok/s
  • GLM-5.312K tok/s
  • gpt-oss-120b51K tok/s
  • Qwen3.6 27B102K tok/s

Cost per million tokens in the inference calculator →

How it compares

SystemArchitectureMemory / GPUBandwidthFP8 denseMax powerCooling
HGX B300Blackwell Ultra288 GB8 TB/s4.5 PF15 kWLiquid / Air
HGX B200Blackwell180 GB8 TB/s4.5 PF14.3 kWAir / Liquid
HGX H200Hopper141 GB4.8 TB/s2.0 PF10.2 kWAir / Liquid
HGX H100Hopper80 GB3.35 TB/s2.0 PF10.2 kWAir / Liquid
HGX A100 80GBAmpere80 GB2 TB/s—6.5 kWAir
RTX Pro 6000 BlackwellBlackwell96 GB1.8 TB/s1.0 PF6.5 kWAir

How it connects to the cluster

Each GPU gets its own network card and its own port on the leaf switch, one rail per GPU. These are the parts on that path.

8× B2008× ConnectX-7 · 400G400G OSFP DR4, flat-top800G OSFP 2×DR4, finnedLeaf: Quantum-2 or Spectrum-4

On this path: ConnectX-7 · 400G OSFP DR4 · 800G OSFP 2×DR4 · Quantum-2 QM9700 · Spectrum-4 SN5600

Alternatives to consider

Consider H100 for immediate capacityH100 is easier to find quickly and draws about 10 kW per node, so it fits more existing data center space.

What to verify before you buy

  • Factory orders: confirm the OEM has acknowledged the order, not just quoted it
  • Delivery window and cancellation terms in writing
  • Power: up to 14.3 kW per node vs 10.2 kW for H100; check rack power before ordering
  • Network: ConnectX-7 400G; confirm switches and optics match an existing NDR fabric

Export controls

B200 carries advanced export controls, stricter than H100/H200. Verify end-user and destination against current BIS entity lists before placing an order.

Frequently asked questions

How much power does an HGX B200 draw?

Up to 14.3 kW per 8-GPU HGX node at maximum rated power. Plan site power for that peak: training runs close to it. Inference typically averages 55–65% of it.

Is the HGX B200 air-cooled or liquid-cooled?

Both. HGX B200 systems ship air or direct-to-chip liquid-cooled. Air racks typically hold 2 nodes (about 29 kW). Direct-to-chip liquid racks typically hold 6 nodes (about 86 kW).

How much GPU memory does an HGX B200 have?

180 GB per GPU, 1,440 GB across eight, at 8 TB/s per GPU.

What models can one HGX B200 node serve?

At FP8 with an 8K context, the largest open models that fit on eight GPUs include DeepSeek-V4.1 Flash, GLM-5.3, gpt-oss-120b, Qwen3.6 27B. The inference calculator shows throughput and cost for each.

What kind of site does an HGX B200 cluster need?

A 32-node cluster (256 GPUs) needs roughly 0.7 MW of facility power with air cooling: a colocation cage or suite.

What comes after the HGX B200?

The HGX B300. The HGX B200 is Blackwell architecture, first shipped in 2024.

Sources: NVIDIA HGX platform · NVIDIA DGX B200. Power is the maximum rated system draw; throughput and site figures are planning estimates from the same models as the Rack Rate tools.

Building with HGX B200?

Send us the cluster spec. We source the full bill of materials, and can finance it.

Start a buildExplore financing