NVIDIA · Data center · Hopper

H100 SXM

Previous generation2022
Specs last checked 21 Nov 2026Report an error

The GPU behind the 2023–24 generative AI build-out.

CUDA cores
16,896
Memory
80GB HBM3
Bandwidth
3,350GB/s
Power
up to 700 W

Overview

Brand
NVIDIA
Segment
Data center
Series
Hopper
Architecture
Hopper
Status
Previous generation
Launch
2022
Launch price (US)
—

GPU

Process
TSMC 4N
Transistors
80 billion
Die size
814 mm²
Streaming multiprocessors
132
CUDA cores
16,896

Memory

Capacity
80 GB
Type
HBM3
Bandwidth
3,350 GB/s

Performance

FP8
~2 PF dense

Board & platform

Power
up to 700 W
NVLink / interconnect
900 GB/s

Variants & versions

H100 PCIe114 SMs, 80 GB HBM2e at about 2 TB/s, 350 W.
H100 NVL94 GB HBM3 at about 3.9 TB/s, PCIe card sold in bridged pairs.
H800 / H20Export-compliant versions for China with reduced interconnect or compute.

Common questions

How much memory does the H100 SXM have?

The H100 SXM has 80 GB of HBM3, giving 3,350 GB/s of bandwidth.

When was the H100 SXM released?

2022.