AWS has just released its most powerful and expensive instance, one that is based on the latest NVIDIA Volta graphics processing units. These new P3 instances are aimed at high performance applications.
Workloads that demand accelerated computation are gaining pace, as companies increasingly turn to machine learning to help propel their businesses. But building new models often requires a great deal of computation.
It’s a good thing, then, that Volta is supposed to be a lot faster than previous generation NVIDIA silicon. And now that it has been made available on the Amazon cloud, company will be able to get started with using it right away.
These EC2 instances are powered by the V100 architecture, which the GPU maker has promised comes with dramatically sped up training of machine learning models.
This makes them perfect for compute intensive tasks like machine learning, deep learning, computational fluid dynamics, computational finance, seismic analysis, molecular modelling, and genomic workloads.
Amazon touts these as the most powerful GPU instances available in the cloud. And the company has made these P3 instances available with 1, 4, or 8 Tesla V100 GPUs, and 8, 32, or 64 custom CPUs based on the Intel Xeon Broadwell processors.
No signs of Skylake yet.
Nevertheless, users get unprecedented power here, with each of the P3 GPU offering access to 5120 CUDA cores and 640 Tensor cores to accelerate the training of deep neural networks. According to Jeff Bar, this instance is 781,000 times faster than the Cray-1 supercomputer that went live in 1976.
Onto the software side of things, AWS has also released a new set of deep learning Amazon Machine Images that include frameworks and tools for building an AI system on AWS.
All this impressive power comes at a cost, however.
In Tokyo, the on-demand rate for the p3.16xlarge instance is $41.94 per hour, which is substantially higher when compared to the $24.67 per hour for the corresponding P2 GPU instance rate.
Then again, no one ever said power came cheap!