NVIDIA GB200 NVL72

rack
72× B200
NVLink
2025
FP64
2.7
PFLOPS
FP32
5.4
PFLOPS
Power
120.0
kW Total
Memory
12960
GB Total

36 GB200 superchips (72 B200 GPUs + 36 Grace CPUs) in a liquid-cooled rack-scale system

FP64
2.66
PFLOPS
27.0 TFLOPS/kW
FP32
5.40
PFLOPS
54.0 TFLOPS/kW
TF32
79.20
PFLOPS
1500.0 TFLOPS/kW
FP16
162.00
PFLOPS
2700.0 TFLOPS/kW

System Details

GPU Configuration

GPU Count: 72 GPUs
Architecture: Blackwell
Interconnect: NVLink

System Specifications

Form Factor: rack
Total Power: 120.0 kW
Total Memory: 12960 GB
Memory Bandwidth: 554400 GB/s

Precision Performance Breakdown

PrecisionSystem PerformancePer GPUEfficiency
FP64 2.664 PFLOPS 37.0 TFLOPS 27.0 TFLOPS/kW
FP32 5.400 PFLOPS 75.0 TFLOPS 54.0 TFLOPS/kW
TF32 79.200 PFLOPS 1100.0 TFLOPS 1500.0 TFLOPS/kW
FP16 162.000 PFLOPS 2250.0 TFLOPS 2700.0 TFLOPS/kW
BF16 162.000 PFLOPS 2250.0 TFLOPS 2700.0 TFLOPS/kW
FP8 324.000 PFLOPS 4500.0 TFLOPS 5400.0 TFLOPS/kW
FP4 648.000 PFLOPS 9000.0 TFLOPS 10800.0 TFLOPS/kW
INT8 324.000 PFLOPS 4500.0 TFLOPS 5400.0 TFLOPS/kW

Powered by NVIDIA B200

This system utilizes 72 × NVIDIA B200 SXM 180GB GPUs, each delivering exceptional performance for AI and HPC workloads.

Per GPU TDP

1000W

Per GPU Memory

180 GB

Process Node

4nm

Architecture

Blackwell

Documentation & Resources

workstation-datasheet-dgx-spark-gtc25-spring-nvidia-us-3716899-web.pdf

NVIDIA • 2025-10-16

View Document ↗

NVIDIA DGX Documentation

System guides, deployment resources

Browse ↗

Typical Use Cases

AI/ML Training
High-Performance Computing
Data Analytics

The NVIDIA GB200 NVL72 with 72× B200 GPUs is designed for enterprise-scale workloads requiring 2.7 PFLOPS of compute power with NVLink interconnect for efficient multi-GPU communication.