New Processors! Graviton3, Trainium, and introducing new instances

On Tuesday, Amazon Web Services announced new EC2 instances running on new versions of its custom-designed chips, the Graviton3 and Trainium. “Making the full power of machine learning available for all customers” is a major goal for AWS CEO Adam Selipsky, who said as much during his keynote address at re:Invent. “Losing money by training and inference are two steps along the road.”

The cloud company, which has its roots in the United States, is now debuting Graviton3 for compute-intensive applications such as high-performance computing, batch processing, electronic design automation (EDA), media encoding, scientific modeling, ad serving, distributed analytics, and CPU-based machine learning inference.

Equipped with the new chip, Equinox2 has more than 25% higher computational performance and twice as much floating-point & cryptographic performance versus Graviton 2. For machine learning, Equinox3 supports bfloat16 data and can provide three times faster performance.

TAWS teased the new C7g instances utilizing Graviton3 processors, which will be released soon. Selipsky also revealed the Trn1 instance, which is powered by Trainium.

For example, AWS introduced Trainium, a CPU designed specifically for training deep learning models, last year. The new servers will provide the greatest price-performance for deep learning training applications like image recognition and natural language processing fraud detection.

The number of instances will be between one and ten. Each instance will have a maximum capacity of 800Gbps for networking bandwidth. Trn1 preview instances are now live.