News Stay informed about the latest enterprise technology news and product updates.

P3 instance lures AI, machine learning workloads

As AI capabilities find uses in new markets, more companies are turning to the cloud for these high-performance computing workloads. And AWS is opening its arms wider with expanded support for GPU-backed instances to provide those resources, at premium prices.

The P3 Elastic Compute Cloud (EC2) instance, released into general availability last week, improves performance for advanced applications with graphics processing units (GPUs). The P3 instance comes in three sizes: p3.2xlarge, p3.8xlarge and p3.16xlarge, with 1-8 NVIDIA Tesla V100 GPUs, 16-128 GBs of GPU memory, 8-64 vCPUs and 61-488 GB of instance memory. The instances also offer enhanced network performance of up to 25Gbps and 14Gbps of Elastic Block Store (EBS) bandwidth.

The P3 instance fits advanced workloads such as machine learning, high performance computing and video processing. It is also one of AWS most expensive instances, ranging from $3.06 to $24.48 per hour for On-Demand pricing.

Amazon also unveiled new Amazon Machine Images (AMIs) for the P3 instance family. These AWS Deep Learning AMIs include frameworks designed specifically for the NVIDIA Volta V100 GPUs included with the P3 instance family. Developers can use the AMIs to build custom AI models and algorithms.

New features and support

  • PostgreSQL compatibility, new features. After months of preview, AWS made PostgreSQL for Amazon Aurora generally available. AWS hopes to entice users to migrate PostgreSQL workloads to Aurora, promising a more scalable, secure and durable managed database service and lower costs. AWS claims PostgreSQL with Aurora has “three times better performance” than standard PostgreSQL databases. Aurora also added the ability to launch R4 instances with a larger cache and faster memory than the previous R3 generation –a developer can double Aurora’s maximum throughput on MySQL databases.
  • New AWS Batch functionality. AWS Batch can now trigger CloudWatch Events when a job transitions from one state to another, so a developer won’t have to poll the state of each Batch job. The event stream feature sends state updates in near real-time, which can route through CloudWatch Events to targets such as AWS Lambda or Amazon Simple Notification Service. AWS also adjusted the service to spin idle EC2 resources down faster in accordance with the cloud provider’s new per-second billing. AWS Batch previously held on to idle resources for the majority of the billing hour to prevent unnecessary instance launches.
  • ElastiCache supports Redis encryption. Redis, an open source in-memory database, does not natively support encryption, but AWS now provides that capability for Amazon ElastiCache. The service now enables encryption for personally identifiable information at rest and in transit. At-rest encryption protects Amazon Simple Storage Service (S3) and disk backups, while in-transit encryption protects data communicated between Redis servers and clients.
  • Apply Glue via CloudFormation. AWS has included its Glue service, which helps execute ETL jobs, as an option for AWS CloudFormation templates. This support helps IT teams automate AWS Glue functions — such as jobs, triggers and crawlers — to quickly load and prepare data for analytics.
  • Address data warehouse demands. Dense compute (DC2) nodes for Amazon Redshift are a second generation of compute clusters designed to reduce latency and boost throughput for demanding data warehouse workloads. The DC2 nodes, which include Intel E5-2686 v4 (Broadwell) CPUs, DDR4 memory and NVMe-based solid state disks, are available for the same price as the previous generation DC1 nodes.
  • Use Elasticsearch in a VPC. Amazon Elasticsearch Service (ES) now supports access from an Amazon Virtual Private Cloud (VPC), which removes the need to connect to the service over the public internet. IT teams can now use Elasticsearch, an open source search engine and analytics service, without configuring firewall rules and domain access policies for ES.
  • Geographic application restriction. AWS Web Application Firewall now includes an option to restrict access to applications based on geographic location to fulfill licensing requirements and security needs. Geographic Match Conditions allows a business to create a whitelist that only allows visitors from specified countries. or a blacklist that blocks access to certain countries.
  • CodePipeline takes pushes from CodeCommit. The latter service can now send an Amazon CloudWatch Event to the former service to trigger a pipeline, which eliminates the need to periodically check for code changes.
  • ALBs support multiple certificates. Businesses can now host multiple secure HTTPS applications and assign each one a Secure Sockets Layer certificate behind one Application Load Balancer. AWS uses Server Name Indication to allow these apps to run on the same load balancer. This means businesses don’t have to use risky Wildcard or complicated Multi-Domain certificates to run multiple HTTPS apps on one load balancer.
  • Migrate to new database sources. The AWS Database Migration Service (DMS) added Azure SQL Database and S3 as sources. S3 was previously supported as a target, but its addition as a source allows teams to freely move data to and from S3 buckets and other DMS sources. Amazon EC2 also now supports Microsoft SQL Server 2017 for extra scalability and performance.