Nvidia releases latest AI chip H200

Nvidia releases its latest AI chip H200, which doubles the inference speed and reduces the cost of use by half

Performance improvements:

H200’s inference speed is almost twice as fast as H100

Memory upgrade:

H200 is the first GPU to use HBM3e memory, providing 141GB of HBM3e (1.4 times), and the memory bandwidth is increased from 3.35TB/second to 4.8TB/second (2 times).

Cost and scale:

The H100 chip costs between US$25,000 and US$40,000. The H200 achieves unprecedented performance while maintaining the same power consumption configuration as the H100, making AI factories and supercomputing systems faster and greener, providing AI and Scientific community brings economic advantage

Compatibility:

H200 is compatible with H100, making it easy for existing users to upgrade.

Based on Hopper architecture:

H200 is based on NVIDIA’s Hopper architecture.

Transformer Engine:

Supports acceleration of large language models and other deep learning models based on the Transformer architecture.

Time to market:

It is planned to be launched in the second quarter of next year, and the production of H100 will be tripled in 2024.

Cloud services and deployment: Systems and cloud instances equipped with H200 will be available starting in the second quarter of 2024. Amazon Web Services, Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure will be the first cloud service providers to offer H200-based cloud instances.

Detailed

More AI News