Skip to main content

AWS launches new GPU instances for machine learning and HPC

AWS
(Image credit: Image Credit: Gil C / Shutterstock)

Amazon Web Services (AWS), the retail giant’s cloud arm, just launched its latest GPU-equipped cloud instances.

Called P4, the new instances will run on Intel Cascade Lake processors and eight Nvidia A100 Tensor Core GPUs.

AWS claims they will have more than double the deep learning performance of the previous generation (which was released almost ten years ago) and will be 60 percent cheaper to train.

At the moment, AWS offers just one size - the p4d.12xlarge instance - which can be launched in the US East (N. Virginia) and US West (Oregon) Regions today. It comes with 320 GB of high-bandwidth GPU memory, as well as 400 Gbps networking, 96 CPU cores, 1.1 TB of system memory and 8 TB of SSD storage.

According to TechCrunch, its on-demand price will be $32.77 per hour, going down to approximately $20 per hour for one-year reserved instances, and $11.57 for three-year reserved instances.

AWS also said that customers can combine 4,000+ GPUs and form an EC2 UltraCluster, which is basically a supercomputer.

P4 instances can now be purchased in On-Demand, Savings Plan, Reserved Instance, and Spot form. Support for use of P4 instances in managed AWS services such as Amazon SageMaker and Amazon Elastic Kubernetes Service will be available “later this year”.