While missing some details, the Arm-based processor has some key new performance metrics
At its annual re:Invent conference, Amazon Web Services announced the newest generation of its Arm-based Graviton processors, the Graviton 3, which the company claims will be 25% or more faster than the last-generation chips in key workloads.
The 25% is likely for integer workloads, because AWS also said the Graviton 3 boasts double the floating-point performances (FLOP), a three-fold performance improvement in machine-learning workloads, and better cryptographic performance. AWS also claims the new chips will use 60% less power.
The chips will power new EC2 C7g instances in the AWS cloud. The chips and instances will be the first to use DDR5 memory, which delivers 50% higher bandwidth than DDR4 but with a much lower power draw.
In announcing the processor, AWS Chief Evangelist Jeff Barr said that they and instances should be useful for compute-intensive workloads such as HPC, batch processing, electronic design automation (EDA), media encoding, scientific modeling, ad serving, distributed analytics, and CPU-based machine-learning inferencing.
AWS yet has to disclose all the technical details about Graviton 3 data-center processor, such as whether it is built on Arm’s Neoverse server cores or uses custom cores designed in house. Amazon did say the new cores support dedicated caches for every vCPU, and support for a new pointer authentication to improve security. The Graviton 3 also supports always-on memory encryption.
The C7g instances are available in preview. It will likely be a few months before general availability.