AWS has announced the launch of its next-generation GPU-equipped instances. The P4d instances arrive nearly a decade after AWS first launched its Cluster GPU instances, which were launched in late 2010, and promise better performance at a lower cost.
“Today I would like to tell you about the new GPU-equipped P4 instances,” Jeff Barr, Chief Evangelist for AWS, said. “These instances are powered by the latest Intel Cascade Lake processors and feature eight of the latest NVIDIA A100 Tensor Core GPUs, each connected to all of the others by NVLink and with support for NVIDIA GPUDirect. With 2.5 PetaFLOPS of floating point performance and 320 GB of high-bandwidth GPU memory, the instances can deliver up to 2.5x the deep learning performance, and up to 60% lower cost to train when compared to P3 instances.”
The new instances come with 1.1 TB of system memory and 8 TB of SSD storage. They are also equipped with four 100 Gbps network connections, 320 GB of high-bandwidth GPU memory and are available for an on-demand price of $32.77 per hour.
The next generation
For large-scale enterprises, there is an opportunity to create an on-demand EC2 UltraCluster by combining 4,000 or more GPUs. These will provide businesses with access to supercomputer-scale machine learning and high-performance computing solutions, including financial modelling, natural language processing and other forms of deep analysis.
Regardless of the size of an organization, the new P4d instances should help businesses solve their biggest machine learning challenges while significantly lowering costs. Productivity benefits are also promised with multi-node distributed training taking less time and more managed services being made available.
With machine learning becoming increasingly prevalent across industries, a broad spectrum of businesses is likely to be interested in AWS’ new GPU-instances, whether they have used the company’s previous instances or not.