AWS is targeting developers of artificial intelligence (AI) applications with new EC2 instances powered by NVIDIA Tesla V100 GPUs.
The new P3 instances deliver 14x performance improvement over P2 instances for machine learning applications, AWS said Wednesday.
The instances are designed for uses including intensive machine learning, deep learning, computational fluid dynamics, computational finance, seismic analysis, molecular modelling, and genomic workloads, according to a blog post by AWS Chief Evangelist Jeff Barr.
The Tesla V100 GPUs are based on NVIDIA’s next-generation Volta architecture, and each have 5,120 CUDA cores and 640 Tensor cores, with speeds up to 125 TFLOPS mixed-precision floating point operations per second. Tensor cores are designed for fast training and inference of large-scale neural networks, and the GPUs will shorten the time necessary to train machine learning models from days to hours, according to the announcement.
The launch means AWS is the first provider to bring Volta-powered instances to market, giving it a potentially significant edge for developers of cutting-edge AI applications. At NVIDIA’s annual conference in May, CEO Jensen Huang said in a keynote address that a “flood” of AI workloads is coming to data centers.
“Comparisons between the P3 and today’s scale-out supercomputers are harder to make, given that you can think of the P3 as a step-and-repeat component of a supercomputer that you can launch on as as-needed basis,” writes Barr.
Running the GPU-powered instances requires CUDA 9 and cuDNN7 drivers and libraries, which are included in the newest Windows AMI versions, and will be in the Amazon Linux AMI update scheduled for November 7.
The new instances are offered in three different sizes, with one, four or eight GPUs with 8, 32 or 64 vCPUs respectively, using Intel Xeon E5 processors, and are now available in four AWS regions; US East, US West, EU West, and Asia Pacific.