AWS unveiled a new entry to its Graviton processor series. The Arm-based Graviton3E is designed for high-performance workloads.
When compared to its predecessor, the processor offers up to 25 percent more general compute performance, 200 percent more floating-point performance and 200 percent more cryptographic performance.
Furthermore, AWS claims that ML workloads are processed up to three times faster. Vector instruction performance reportedly rose by up to 35 percent. All in all, Graviton3E chips are a great fit for high-performance computing workloads.
AWS is deploying Graviton3E processors in its new EC2 Hpc7G instances. The series is expected to become available in 2023. The instances will feature various configurations with up to 64 vCPUs and 128 gibibytes (GiB) of memory.
AWS is also incorporating the new processors into its recently announced EC2 c7gn Graviton3E instances. These instances offer network speeds of up to 200 Gbps and up to 50 percent more packet processing performance than existing Graviton instances.
The series is suitable for network-intensive workloads such as virtual network appliances, data analytics and processor-heavy AI and ML inference applications.
All new instances run on the tech giant’s new Nitro System Cards. The cards deliver the fifth version of AWS’ Nitro hypervisor. This version should offer less latency, up to 40 percent more performance per watt and 60 percent higher PPS.
Techzine is attending AWS re:Invent this week. Keep an eye on the website for more. You can find the most important news in this overview.