Heroic Web services are ready to announce updates of their Graviton4 chip, including 600 gigabytes per second of network bandwidth, which the company calls the highest offering in public clouds.
Ali Saidi, an iconic engineer at AWS, compared the speed of a second -reading machine to 100 music CDs.
Graviton4, a central processing unit, or CPU, is one of several chip products that come from Annapurna Labs of Amazon in Austin, Texas. Chip is a win for the company’s custom strategy and puts it against traditional semiconductor players like Intel And AMD,
But the real fight is with Nvidia In artificial intelligence Infrastructure Space.
AWS’s Re: Invented in the invention 2024 conference last December, the company announced the project renier – an AI super computer manufactured for startups anthropicAWS has put $ 8 billion in backing anthropic.
AWS Senior Director Gadi Hut for customer and project engineering said that Amazon AI is looking to reduce training costs and provide an option for NVidia’s expensive graphics processing units or GPU.
According to AWS, the cloud OPUS 4 AI model of Anthropic was launched at Trainium2 GPU, and the project rainier is operated by more than half a million chips – an order that would have traditionally moved into NVIDIA.
Hut said that while Nvidia’s Blackwell is a high -performing chip compared to Trainium 2, the AWS chip offers better cost performance.
He said, “Trainium 3 is coming this year, and it is double the performance of Trainyum 2, and it is going to save energy from the additional 50%,” he said.
According to Rami Sino, director of engineering at AWS ‘Annapurna Labs, the demand for these chips is already overtaking the supply.
“Our supply is very, very large, but there is a customer in every service we created by us,” he said.
With upgrade on the horizon of gravitone 4 and the trainum chips of the project rainier, from Amazon Networking to training, the entire AI is demonstrating its widespread ambitions to control the infrastructure stack.
And more prominent AI models such as Cloud 4 prove to be successfully trained on non-NVidia hardware, the question is not whether the AWS chip can compete with the giant-how much market share it can take.
According to AWS spokesperson, the release schedule for Graviton4 updates will be provided by the end of June.
