August 9, 2017 Timothy Prickett Morgan
One of the luckiest coincidences in the past decade has been that the hybrid machines designed for traditional HPC simulation and modeling workloads. which combine the serial processing performance of CPUs and the parallel processing and massive memory bandwidth of GPUs, were also well suited to run machine learning training applications.
If the HPC community had not made the investments in hybrid architectures, the hyperscalers and their massive machine learning operations, which drive just about all aspects of their businesses these days, would not have seen such stellar results. (And had that not happened, many of us would have had …Read more