In January 2022, we launched Amazon EC2 Hpc6a situations for patrons to effectively run their compute-bound excessive efficiency computing (HPC) workloads on AWS with as much as 65 % higher worth efficiency over comparable x86-based compute-optimized situations.
As their jobs develop extra advanced, clients have requested for extra cores with extra compute efficiency and extra reminiscence and community efficiency to scale back the time to finish jobs. Moreover, as clients look to convey extra of their HPC workloads to EC2, they’ve requested how we are able to make it simpler to distribute processes to make the very best use of reminiscence and community bandwidth, to align with their workload necessities.
As we speak, we’re asserting the overall availability of Amazon EC2 Hpc7a situations, the following technology of occasion varieties which are purpose-built for tightly coupled HPC workloads. Hpc7a situations powered by the 4th Gen AMD EPYC processors (Genoa) ship as much as 2.5 occasions higher efficiency in comparison with Hpc6a situations. These situations provide 300 Gbps Elastic Cloth Adapter (EFA) bandwidth powered by the AWS Nitro System, for quick and low-latency internode communications.
Hpc7a situations characteristic Double Information Price 5 (DDR5) reminiscence, which offers 50 % larger reminiscence bandwidth in comparison with DDR4 reminiscence to allow high-speed entry to knowledge in reminiscence. These situations are perfect for compute-intensive, latency-sensitive workloads similar to computational fluid dynamics (CFD) and numerical weather prediction (NWP).
If you’re operating on Hpc6a, you need to use Hpc7a situations and benefit from the two occasions larger core density, 2.1 occasions larger efficient reminiscence bandwidth, and three occasions larger community bandwidth to decrease the time wanted to finish jobs in comparison with Hpc6a situations.
Right here’s a fast infographic that reveals you the way the Hpc7a situations and the 4th Gen AMD EPYC processor (Genoa) evaluate to the earlier situations and processor:
Hpc7a situations characteristic sizes of as much as 192 cores of the AMD EPYC processors CPUs with 768 GiB RAM. Listed below are the detailed specs:
|Occasion Title||CPUs||RAM (Gib)
||EFA Community Bandwidth (Gbps)
||Hooked up Storage|
|Hpc7a.12xlarge||24||768||As much as 300||EBS Solely|
|Hpc7a.24xlarge||48||768||As much as 300||EBS Solely|
|Hpc7a.48xlarge||96||768||As much as 300||EBS Solely|
|Hpc7a.96xlarge||192||768||As much as 300||EBS Solely|
These situations present larger compute, reminiscence, and community efficiency to run probably the most compute-intensive workloads, similar to CFD, climate forecasting, molecular dynamics, and computational chemistry on AWS.
Much like EC2 Hpc7g situations launched a month earlier, we’re providing smaller occasion sizes that makes it simpler for patrons to choose a smaller variety of CPU cores to activate whereas preserving all different assets fixed primarily based on their workload necessities. For HPC workloads, widespread eventualities embrace offering extra reminiscence bandwidth per core for CFD workloads, allocating fewer cores in license-bound eventualities, and supporting extra reminiscence per core. To be taught extra, see Occasion sizes within the Amazon EC2 Hpc7 household – a unique expertise within the AWS HPC Weblog.
As with Hpc6a situations, you need to use the Hpc7a occasion to run your largest and most advanced HPC simulations on EC2 and optimize for value and efficiency. You can even use the brand new Hpc7a situations with AWS Batch and AWS ParallelCluster to simplify workload submission and cluster creation. You can even use Amazon FSx for Lustre for submillisecond latencies and as much as lots of of gigabytes per second of throughput for storage.
To attain the very best efficiency for HPC workloads, these situations have Simultaneous Multithreading (SMT) disabled, they’re obtainable in a single Availability Zone, they usually have restricted exterior community and EBS bandwidth.
Amazon EC2 Hpc7a situations can be found at present in three AWS Areas: US East (Ohio), EU (Eire), and US GovCloud for buy in On-Demand, Reserved Situations, and Financial savings Plans. For extra info, see the Amazon EC2 pricing web page.