NVIDIA HGX H100 4-GPU

baseboard
4× H100
NVLink
2022
FP64
0.1
PFLOPS
FP32
0.3
PFLOPS
Power
4.2
kW Total
Memory
320
GB Total

HGX H100 4-GPU configuration for smaller deployments

FP64
0.14
PFLOPS
32.4 TFLOPS/kW
FP32
0.27
PFLOPS
63.8 TFLOPS/kW
TF32
1.98
PFLOPS
941.9 TFLOPS/kW
FP16
3.96
PFLOPS
1884.8 TFLOPS/kW

System Details

GPU Configuration

GPU Count: 4 GPUs
Architecture: Hopper
Interconnect: NVLink

System Specifications

Form Factor: baseboard
Total Power: 4.2 kW
Total Memory: 320 GB
Memory Bandwidth: 13400 GB/s

Precision Performance Breakdown

PrecisionSystem PerformancePer GPUEfficiency
FP64 0.136 PFLOPS 34.0 TFLOPS 32.4 TFLOPS/kW
FP32 0.268 PFLOPS 67.0 TFLOPS 63.8 TFLOPS/kW
TF32 1.978 PFLOPS 494.5 TFLOPS 941.9 TFLOPS/kW
FP16 3.958 PFLOPS 989.5 TFLOPS 1884.8 TFLOPS/kW
BF16 3.958 PFLOPS 989.5 TFLOPS 1884.8 TFLOPS/kW
FP8 7.916 PFLOPS 1979.0 TFLOPS 3769.5 TFLOPS/kW
INT8 7.916 PFLOPS 1979.0 TFLOPS 3769.5 TFLOPS/kW

Powered by NVIDIA H100

This system utilizes 4 × NVIDIA H100 SXM5 80GB GPUs, each delivering exceptional performance for AI and HPC workloads.

Per GPU TDP

700W

Per GPU Memory

80 GB

Process Node

4nm

Architecture

Hopper

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

Large Language Model Training
Distributed Deep Learning
Multi-GPU Inference
HPC Simulations
Scientific Computing

The NVIDIA HGX H100 4-GPU with 4× H100 GPUs is designed for enterprise-scale workloads requiring 0.1 PFLOPS of compute power with NVLink interconnect for efficient multi-GPU communication.