Machine Learning and Deep Learning
NEC HPC Platform for Machine Learning and Deep Learning
HPC capabilities have been used for scientific simulations to pursue higher simulation resolution, larger simu-lation scale, and therefore more accurate scientific result. Those HPC capabilities have contributed to scientific advancements through computation sciences. They have now started to be used for emerging technologies like Artificial Intelligence (A.I.), Machine Learning (ML), Deep Learning (DL), as well, with Big Data Analytics a vital ingredient of it all.
ML and DL have been used to empower business, research, and development competitiveness. These methods make suggestions and even decisions by processing huge amounts of data, therefore typical HPC capabilities are needed to get results faster, to handle larger data sets, and to produce results that are more accurate.
There are several applications in the ML and DL field, and those have different algorithmic characteristics. In the ML field, Apache Spark or Scikit-learn are widely used, often in applications that require a very high memory bandwidth. On the other hand, frameworks like TensorFlow or Pytorch are very famous in the DL field. Those applications require higher peak performance rather than memory bandwidth.
The Vector Engine processor of NEC SX-Aurora TSUBASA supercomputer excels in memory bandwidth because of an elaborate memory architecture based on six HBM2-modules per CPU. Due to these characteristics of the CPU, such a Vector Engine can accelerate applications of the ML field and a portion of the DL field.