AWS has just announced its new Graviton4 processor, the fourth-generation Arm-based server CPU that has 50% more cores, 75% more memory bandwidth, and an overall compute performance uplift of up to 30%.

Amazon Web Services (AWS) was previously manufacturing its Graviton3 and Graviton3E processors on the 5nm process node at Taiwan Semiconductor Manufacturing Company (TSMC) but is moving to the newer, mature 4nm process node that NVIDIA is using to build its Grace CPU and "Hopper" H100 and H200 AI GPUs.
Inside, the new Graviton4 has 96 cores based on Arm's "Demeter" Neoverse V2 core (which is based on the Armv9 architecture), which is a 50% boost in cores over Graviton3 which had 64 cores.
There are also 12 x DDR5 controllers on Graviton4, compared to 8 x DDR5 controllers on Graviton3. This means we have more memory bandwidth: Graviton4 features 536.7GB/sec of memory bandwidth per socket, compared to 307.2GB/sec of memory bandwidth per socket on Graviton3 and Graviton3E, a huge 75% uplift in memory bandwidth there.
- Read more: Amazon announces Q just days after ChatGPT maker OpenAI announces Q*
- Read more: Amazon AWS Graviton3 server CPU: triple-socket motherboard support
AWS compute and networking VP David Brown said in a press release: "Graviton4 marks the fourth generation we've delivered in just five years and is the most powerful and energy-efficient chip we have ever built for a broad range of workloads. By focusing our chip designs on real workloads that matter to customers, we're able to deliver the most advanced cloud infrastructure to them".

The new Graviton4 process will be available in Amazon EC2 R8g instances, which are available right now in preview, with general availability to happen over the next few months.