AWS’s new server processor can train AI models up to four times faster

The two processors are designed to deliver supercomputer-level performance

Key Takeaways

  • Amazon announces development of new processors, Graviton4 and Trainium2, for AWS-exclusive work, with a focus on speed and performance.
  • Graviton4 is the most powerful AWS processor, enhancing compute performance by up to 30% and offering 75% more memory bandwidth than its predecessor.
  • Trainium2 processor is designed for AI training, four times faster than competitors, with triple the memory capacity and improved energy-efficiency.

As Amazon Web Services (AWS) continues to grow, Amazon’s product line is expanding, too. Now, the company has announced that it has developed two new processors, which will be used by AWS. While one of them was created for general-purpose work, the other was made with AI training in mind — and Amazon specifies that speed is the name of the game.

On Nov. 28, Amazon announced the development of two new chips — the AWS Graviton4 and AWS Trainium2 — for AWS-exclusive work. The company says that the Graviton4 is currently the most powerful AWS processor in its lineup, with the ability to enhance compute performance by up to 30%. It also has 75% more memory bandwidth than Graviton3 processors, and it will be available in will be available in Amazon EC2 R8g instances — this can help with tasks involving high-performance databases, as well as large data analytics workloads. Tom’s Hardware specifies that the Graviton4 features Arm’s 64-bit architecture and includes 96 cores. However, Amazon did not mention whether the chip specifically uses Arm’s Neoverse V2 cores.

The Trainium2 processor from Amazon was developed specifically for AI training, and the company says it’s four times faster than its competitors. It also has triple the memory capacity of its predecessor and can achieve up to twice the performance per watt, making it significantly more energy-efficient. Amazon noted that machines using the processor will connect via the AWS Elastic Fabric Adapter — a network interface for high performance computing — and deliver petabit-scale performance. Trainium2 was created to cut down on the training time associated with foundation models and large language models, which can take months.

Although many people don’t immediately associate Amazon with AI, the company is continuing its push into the industry. As it works to develop more robust processors, it’s hard to imagine that Amazon’s venture into the market will slow any time soon.

(Visited 5 times, 1 visits today)

Related posts