Interactive 3D Model

Loading viewer...

Drag to rotate ┬╖ Scroll to zoom

ARCHITECTURE

Hopper

SERIES

GH200

Memory

96GB HBM3 + 480GB LPDDR5X

Bandwidth

—

TDP

—

On OrderGH200┬╖Best for: HPC & AI

NVIDIA GH200 Grace Hopper

Hopper

|Data Center GPUs

Overview

The NVIDIA GH200 Grace Hopper Superchip combines the Grace CPU with the Hopper GPU via NVLink-C2C for groundbreaking HPC and AI performance. Ideal for supercomputing clusters and large-scale scientific simulations.

Key Features

144 Arm Cores
96GB HBM3 GPU
NVLink-C2C

Ideal Use Cases

hpcai training

Full Specifications

Complete technical specifications for the GH200.

Memory

Memory: 96GB HBM3 + 480GB LPDDR5X
Memory Bandwidth: —

Compute

Tensor Cores: —
CUDA Cores: —
FP8 TFLOPS: —
FP16 TFLOPS: —
TF32 TFLOPS: —
FP64 TFLOPS: —

Interconnect

Interconnect: NVLink-C2C, 900GB/s

Physical

TDP: —
Form Factor: Module
Cooling: Liquid Cooled

Manufacturing

Launch Date: Q2 2024
Manufacturing Process: 4nm

Ready to deploy the GH200?

Get a personalized quote within 24 hours. Our NVIDIA-certified engineers will help you architect the optimal configuration for your workload.

Related Chips

More chips in Data Center GPUs.

H100

In Stock

NVIDIA H100 Tensor Core GPU

Enterprise AI training and inference GPU with Transformer Engine and FP8 precision.

ArchitectureHopper

Memory80GB HBM3

TDP700W

H200

In Stock

NVIDIA H200 Tensor Core GPU

Next-gen AI GPU with 141GB HBM3e memory for larger models and faster inference.

ArchitectureHopper

Memory141GB HBM3e

TDP700W

B200

Pre-Order

NVIDIA B200 Tensor Core GPU

Flagship AI GPU with Blackwell architecture, 384GB HBM3e, and 10 TB/s bandwidth.

ArchitectureBlackwell

Memory384GB HBM3e

TDP1000W

A100

In Stock

NVIDIA A100 Tensor Core GPU

Proven enterprise GPU for AI, HPC, and data analytics workloads.

ArchitectureAmpere

Memory80GB HBM2e

TDP400W