At the recent re:Invent conference, Amazon introduced new EC2 instance families running on different processors, including G5g Arm-based instances, M6a AMD-based ones and storage-optimized Im4gn and Is4gen.
Part of the M6 "general purpose" portfolio, the M6a instances are designed to provide a balance of compute, memory, storage, and network resources. They feature 3rd generation AMD EPYC processors, running at frequencies up to 3.6 GHz, and are supposed to offer a 35% price performance improvement on previous generation M5a instances. The M6a instances follow the previous M6 announcements of M6g (Arm-based) instances and M6i (x86-based). Channy Yun, principal developer advocate at AWS, highlights one of the benefits of the M6a:
The new M6a instances offer larger instance size with 48xlarge with up to 192 vCPUs and 768 GiB of memory, enabling you to consolidate more workloads on a single instance. M6a also offers Elastic Fabric Adapter (EFA) support for workloads that benefit from lower network latency and highly scalable inter-node communication.
The cloud provider has also announced the general availability of the G5g family, the first hardware-accelerated instances running on Graviton2 and targeting GPU-based workloads, such as streaming Android gaming, ML inference, graphics rendering and autonomous vehicle simulations. The G5g instances run NVIDIA T4G Tensor Core GPUs with up to 25 Gbps of networking bandwidth and 19 Gbps of EBS bandwidth.
The Im4gn and Is4gen instances are the latest storage-optimized types designed for applications requiring high throughput and low-latency access to large amounts of data on local SSD storage. Common use cases are SQL and NoSQL databases running on EC2.
Providing lower cost per TB and higher density per vCPU of SSD storage, the Im4gn instances are designed for applications that are not memory intensive but require high compute performance, for example social games, session storage, chatbots, and search engines. With up to 100 Gbps of networking bandwidth, the Is4gen instances benefit applications performing large amounts of random I/O.
In the article Scaling Kafka at Honeycomb, Liz Fong-Jones, principal developer advocate for SRE & observability at Honeycomb, writes:
By giving us the right ratio of compute, SSD, and network throughput, Im4gn is the most attractive instance family to us to sustain our long-term growth. In the 4xlarge size, it has guaranteed (e.g., not just burst w/ half that as steady-state) 25 Gbps of bandwidth, plenty of CPU (utilizing its 16 more efficient Graviton2 cores), and 7.5T of instance storage.
All the new instance types are already generally available, even if some only in a subset of regions. On-demand pricing depends on instance type, size and region.