Amazon The web service is configured to announce updates to the Graviton4 chip, which includes 600 Gigabits per second of network bandwidth.
Ali Saidi, a well-known AWS engineer, compared speed to a machine that reads 100 music CDs per second.
The central processing unit or CPU, Graviton4, is one of many chip products from Amazon’s Annapurna Labs in Austin, Texas. This chip is a victory in the company’s custom strategy and is to compete against traditional semiconductor players. Intel and AMD.
But there’s a real battle nvidia In the artificial intelligence infrastructure space.
AWS RE: At the Invention 2024 conference last December, the company announced Project Rainier, an AI supercomputer built for startup humanity. AWS has put $8 billion in backing humanity.
Gadi Hutt, senior director of AWS customer and product engineering, said Amazon is trying to reduce AI training costs and provide an alternative to Nvidia’s expensive graphics processing units (GPUs).
According to AWS, the human Claude Opus 4 AI model will be released on the Trainium2 GPU, with the Rainier carrying over 500,000 chips.
Hutt said Nvidia’s Blackwell is a better chip than Trainium2, but AWS chips offer better cost performance.
“This year we’re introducing Trainium 3, double the performance of Trainium 2, saving an additional 50% energy,” he said.
Demand for these chips is already outweighing supply, according to Rami Sinno, engineering director at Annapurna Labs at AWS.
“Our supply is very, very large, but every service we build has customers attached,” he said.
With the Graviton4 upgrades with Horizon and Project Rainier’s Trainium Chips, Amazon demonstrates its broader ambitions to control the entire AI infrastructure stack, from networking to training.
Also, it doesn’t matter whether AWS can compete with chip giants, as it proves that major AI models like Claude 4 can train well on non-Nvidia hardware.
According to an AWS spokesperson, the release schedule for the Graviton 4 update will be available by the end of June.
Fix: Previous versions of this story misunderstood the nature of the bandwidth of the updated Graviton 4 chip.
