Japanese computer giant NEC Corp. claims to have developed new data processing technology capable of accelerating machine learning on vector computers by up to 50 times that of the popular Apache Spark big data framework.

The company said its new technology leverages something called “sparse matrix” data structures to significantly boost the performance of vector computers in machine learning tasks.

Vector computers are basically supercomputers that are built primarily to handle large scientific and engineering calculations. Vector computer architecture was the most dominant in supercomputing through the 1970s to 1990s, used in the majority of legendary Cray Inc.’s platforms during that period. The architecture is far less common now because of the superior performance of conventional microprocessor designs, but vector has been given a new lease of life by NEC in recent years with the development of its SX-series of supercomputers.

NEC said it has also developed new middleware alongside its data processing technology that incorporates sparse matrix structures to simplify the use of machine learning. The company said this middleware can be launched from Python and Spark infrastructures without special programming.

Regarding NEC’s claim that its technology is 50 times faster than Apache Spark, this is to be expected as it’s essentially comparing apples to oranges, said Holger Mueller, vice president and principal analyst at Constellation Research Inc. He explained that Spark is often used as in-memory storage to feed data into neural networks, which often run on graphics processing units from the likes of Nvidia Corp. Meanwhile, vector computers are optimized to compute and process vectors, which are the basic unit of a neural network.

“There’s no surprise that it is faster at modeling, creating and parsing a neural network. This is why Nvidia is doing so well with artificial intelligence and neural networks as its GPUs are essentially vector calculators,” Mueller said. “How fast would a combination of Spark for storage and NEC’s vector computers for parsing data, learning and computing be? Probably the fastest, but we need to learn more from NEC first.”

Meanwhile, NEC was keen to emphasize its new platform’s cost-effectiveness and possible use cases.

“This technology enables users to quickly benefit from the results of machine learning, including the optimized placement of web advertisements, recommendations, and document analysis,” said Yuichi Nakamura, general manager of system platform research laboratories at NEC. “Furthermore, low-cost analysis using a small number of servers enables a wide range of users to take advantage of large-scale data analysis that was formerly only available to large companies.”

NEC added that its latest SX-ACE vector computer is designed to meet a wide range of performance needs, and that the new data processing technology expands its capabilities to include large-scale data analysis such as machine learning and deep learning.

The company said it’s demonstrating its new technology at the International Symposium on Parallel and Distributed Computing 2017 in Innsbruck, Austria, that runs through Thursday.

Image: Orlando@Roll/Flickr