AWS Levels Up Its Workhorse Chips, Graviton and Trainium
Amazon Web Services has updated its Graviton and Trainium chips with an eye toward power and efficiency, the company announced Tuesday at its re:Invent conference.
The new Trainium2 chip, ideal for training AI models, is now up to four times faster and two times more efficient than its predecessor, with three times the memory. In its new iteration, the chip can support training foundational and large learning models "with up to trillions of parameters," according to AWS.
"Trainium2 will be available in Amazon EC2 Trn2 instances, containing 16 Trainium chips in a single instance," the company explained. "Trn2 instances are intended to enable customers to scale up to 100,000 Trainium2 chips in next generation EC2 UltraClusters, interconnected with AWS Elastic Fabric Adapter (EFA) petabit-scale networking, delivering up to 65 exaflops of compute and giving customers on-demand access to supercomputer-class performance. With this level of scale, customers can train a 300-billion parameter LLM in weeks versus months."
The other new release, Graviton4, is the most powerful version of the general-purpose chip. Compared to its predecessor, Graviton4 has 75 percent more memory, 50 percent more cores and a 30 percent performance improvement.
As organizations mature and grow in the cloud, so has the size of their data and the complexity of their workloads. AWS positions Graviton4 as the ideal silicon to support these changes, while still keeping costs and energy consumption down.
"Graviton4 will be available in memory-optimized Amazon EC2 R8g instances, enabling customers to improve the execution of their high-performance databases, in-memory caches, and big data analytics workloads," AWS said. "R8g instances offer larger instance sizes with up to 3x more vCPUs and 3x more memory than current generation R7g instances. This allows customers to process larger amounts of data, scale their workloads, improve time-to-results, and lower their total cost of ownership."
More information on the Graviton4 is available here and on the Trainium2 here.