No datasheet is linked because NVIDIA does not publish a product page for this part. It is an export-compliance variant sold into the Chinese market, and the addresses that would hold it return 404. Specifications for these variants circulate through server makers and resellers rather than through NVIDIA.

NVIDIA H20 96GB

Hopper SXM 2024 4nm Spec confidence: Official
FP8 (dense)
296
TFLOPS
FP32
44
TFLOPS
VRAM
96 GB
HBM3
Bandwidth
4.0 TB/s
memory
TDP
400 W
110.0 TFLOPS/kW
Download datasheet

Overview

The NVIDIA H20 is a Hopper architecture datacenter GPU designed for the Chinese market under U.S. export regulations. Featuring 96GB HBM3 memory with 4.0 TB/s bandwidth, it delivers 44 TFLOPS FP32 and 74 TFLOPS TF32 at 400W TDP. While compute is constrained compared to the H100, its exceptional memory bandwidth suits large language model inference.

Performance

Peak theoretical throughput by precision type

INT8
8-bit integer
296 TOPS
sparse not published
FP8
8-bit floating point
296 TFLOPS
592 with sparsity
FP16
16-bit floating point
148 TFLOPS
sparse not published
BF16
Brain Float 16
148 TFLOPS
sparse not published
TF32
TensorFloat-32
74 TFLOPS
sparse not published
FP32
32-bit floating point
44 TFLOPS
FP64
64-bit floating point
1 TFLOPS
Dense Added with sparsity
0.740 TFLOPS/W FP8 at 400 W
0.370 TFLOPS/W FP16 at 400 W
0.110 TFLOPS/W FP32 at 400 W

The H20 96GB in the GPU landscape

Peak FP16 TFLOPS (dense) against TDP, single-GPU parts tracked by Flopper

06001,2001,8002,4003,0000 W250 W500 W750 W1000 W1250 W1500 WInstinct MI355XH20 96GB

Higher and further left is better: more half-precision throughput for less power.

Specifications

Architecture

Hopper

Form Factor

SXM

Launch Year

2024

Process Node

4nm

Memory

96 GB HBM3

Bandwidth

4,000 GB/s

TDP

400 W

Max Power

~460 W

Spec Confidence

Official

Full Specifications

Memory
VRAM 96 GB
Memory Type HBM3
Bandwidth 4.0 TB/s
Power & Thermal
TDP 400 W
Enterprise Features
Sparsity Yes
General
Form Factor SXM
Architecture Hopper
Launch Year 2024

Datasheet & Resources

Flopper Datasheet

NVIDIA H20 96GB specifications, generated from our database. Printable.

NVIDIA Developer Documentation

Technical specs, programming guides

Browse

Data Provenance

Every figure traced to a source

Primary Source

Document
--
Publisher
NVIDIA
Published
--

Data Quality

Spec confidence
Official
Clock basis
Boost
Precisions tracked
7
Normalization
All values in TFLOPS

Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.

Browse GPU Cloud Providers

Similar GPUs

NVIDIA H100 SXM5 80GB

SXM · 2022
FP32: 67 TFLOPS
Compare vs H100

NVIDIA H200 SXM 141GB

SXM · 2024
FP32: 67 TFLOPS
Compare vs H200

NVIDIA GH200 96GB HBM3

SXM · 2023
FP32: 67 TFLOPS
Compare vs GH200

NVIDIA GH200 144GB HBM3e

SXM · 2024
FP32: 67 TFLOPS
Compare vs GH200

NVIDIA H100 NVL 94GB

PCIe · 2023
FP32: 60 TFLOPS
Compare vs H100

Frequently Asked Questions

How many TFLOPS does the NVIDIA H20 have?

The NVIDIA H20 delivers 44 TFLOPS for FP32 operations, 148 TFLOPS for FP16, and 296 TFLOPS for FP8 precision.

What is the power consumption of the NVIDIA H20?

The NVIDIA H20 has a TDP (Thermal Design Power) rating of 400 watts.

How much memory does the NVIDIA H20 have?

The NVIDIA H20 is equipped with 96 GB of VRAM with 4000 GB/s memory bandwidth.

What architecture is the NVIDIA H20 based on?

The NVIDIA H20 is based on the Hopper architecture, launched in 2024.

Stay Updated on GPU Releases

Get notified when new GPUs are added or specifications are updated.

Loading verification...

No spam, unsubscribe anytime.

Back to GPUs