AWS rolls out Graviton2-powered EC2 instances for GPU-based workloads
Amazon Web Services on Monday said it's bringing a new set of EC2 instances into general availability, including Graviton2-based instances designed for GPU-based workloads. Along with Amazon's custom-designed Graviton2 chip, the Amazon EC2 G5g instances also feature Nvidia T4G Tensor Core GPUs.
AWS highlighted a few workloads that G5g instances would serve well: For Android game streaming, the instances provide up to 30% lower cost per stream per hour than x86-based GPU instances, Amazon said. For ML inference, G5g instances are well-suited for models that are sensitive to CPU performance or leverage Nvidia's AI libraries. For graphics rendering, G5g instances are the most cost-effective option for AWS customers.
The instances are compatible with a number of graphical and machine learning libraries on Linux, including NVENC, NVDEC, nvJPEG, OpenGL, Vulkan, CUDA, CuDNN, CuBLAS, and TensorRT.
The G5g instances are currently available in the US East (N. Virginia), US West (Oregon), and Asia-Pacific (Seoul, Singapore and Tokyo) Regions in On-Demand, Spot, Savings Plan, and Reserved Instance form.
Also on Monday, Amazon announced the AWS Graviton Ready Program, which validates software from partners to ensure customers have a consistent experience using software on Graviton as they do on other instances.
AWS is also bringing into GA the EC2 M6a instances featuring the 3rd Gen AMD Epyc processors. Running at frequencies up to 3.6
GHz, they offer up to 35% price performance versus the previous generation M5a instances. These new instances are well-suited for general-purpose workloads such as web servers, application servers and small data stores.
In addition to providing better price-performance than M5a instances, the M6a instances come in larger sizes, with 48xlarge with up to 192 vCPUs and 768 GiB of memory. M6a also offers Elastic Fabric Adapter (EFA) support for workloads that benefit from lower network latency and highly scalable inter-node communication, such as HPC or video processing.
M6a instances are currently available in 10 sizes in the AWS US East (N. Virginia), US West (Oregon), and Europe (Ireland) Regions as On-Demand, Spot, and Reserved Instance or as part of a Savings Plan.