The Amazon could computing division Amazon Web Services has just announced a new Elastic Compute Cloud (EC2) instance type. Dubbed P2, this new Amazon cloud service is taking advantage of Nvidia high-performance GPUs in order to offer massive computing performance.
Amazon Web Services Powered By Nvidia GPUs
According to Hot Hardware, Nvidia GPUs help Amazon avoid the inherent limitation and bottlenecks of the current CPUs and architecture. By adding Nvidia GPUs in its cloud architecture, Amazon is able to enhance the performance of its cloud computing services.
Jeff Barr, Amazon's Chief Evangelist at Amazon Web Services, explains in a blog post on the AWS website that the GPU scales out instead of up. This means that the graphics processing units offer parallel banks of memory and more processors instead of bottlenecked memory and faster processors.
According to Forbes, for Amazon the net effect of leveraging Nvidia GPUs is that massively parallel workloads can be processed more efficiently and significantly faster than they could on CPUs. Amazon Web Services has designed the new AWS P2 Instances for deep learning, large-scale machine learning, seismic analysis, computational fluid dynamics, genomics, molecular modeling, and computational finance workloads.
The new Amazon cloud computing P2 instances incorporate up to eight Tesla K80 accelerators from Nvidia. Each P2 instance is running a pair of Nvidia GK210 GPUs.
According to ExtremeTech, the Nvidia Tesla K80 accelerators are based on the older Kepler architecture that is previous to the Maxwell architecture available since 2014. The older Maxwell architecture was explicitly designed as a workstation and consumer product.
The GK210 in the Tesla K80 offers much more shared memory and a larger register file per multiprocessor block. The K80 GPUs in the new P2 instances will come with 2,496 CUDA cores and 12 GB of RAM per card. All K80s offer up to 240GB/s of memory bandwidth per card and support ECC memory protection.