H100 CNX

GPU Code: GH100

80 GB

HBM2e

1845 MHz

Boost Clock

350 W

TDP

5 nm

Process
Share on Social Media
Share H100 CNX specifications with others

Technical Specifications

Core Specifications
Architecture: Hopper
Process Size: 5 nm
Base Clock: 690 MHz
Boost Clock: 1845 MHz
Shader Units: 14592
SMs: 114
Tensor Cores: 456
TMUs: 456
ROPs: 24
Memory & Performance
Memory Size: 80 GB
Memory Type: HBM2e
Memory Bandwidth: 2040.0 GB/s
Bus Interface: PCIe 5.0 x16
TDP: 350 W
Suggested PSU: 750 W
Cache & Die Information
L1 Cache: 250 KB
L2 Cache: 50.0 MB
Die Size: 814 mm²
Transistors: 80.0 billion
Transistor Density: 98.3 million/mm²

API Support

Cuda Compute
9.0 sm_90

Physical Specifications

Board Length: 267 mm
Board Width: 111 mm
Power Connectors: 8-pin EPS

Quick Facts

Release Date: 2023-03-21
Manufacturer: NVIDIA
Generation: Server Hopper(Hxx)

Performance

Pixel Fill Rate 44.3 GPixel/s
Texture Fill Rate 841.3 GTexel/s
FP32 Performance 53.8 TFLOPS

CUDA Toolkit Compatibility

Cuda Compute
9.0 sm_90
Architecture

Hopper

Supported CUDA Toolkits
11.8 12.0 12.1 12.2 12.3 12.4 12.5 12.6 12.7 12.8

Hopper architecture with Transformer Engine support for AI workloads