NEC is known for its vector processor-powered supercomputers, most notably the Earth Simulator. Typically, NEC’s vector processors have been aimed at numerical simulation and similar workloads, but recently NEC unveiled a platform that makes its latest SX-Aurora Tsubasa supercomputer-class processors usable for artificial intelligence and machine learning workloads.
“The vector processor, with advanced pipelining, is a technology that proved itself long ago,” wrote Robbert Emery, who is responsible for commercializing NEC Corporation’s advanced technologies in HPC and AI/ML platform solutions.
“Vector processing paired with middleware optimized for parallel pipelining is lowering the entry barriers for new AI and ML applications, and is set to solve the challenges both today and in the future that were once only attainable by the hyperscale cloud providers.”
NEC’s SX-Aurora for AI
The SX-Aurora Tsubasa AI Platform supports both Python and TensorFlow development environments as well programming languages such as C/C++ and Fortran.
NEC offers multiple versions of its latest SX-Aurora Tsubasa versions for desktops and servers that can handle FHFL cards. The most advanced Vector Engine Processor model is the Type 20 that features 10 cores running at 1.6GHz and paired with 48GB HBM2 memory. The card offers a peak performance of 3.07 FP32 TFLOPS or 6.14 FP16 TFLOPS.
While peak performance numbers offered by the SX-Aurora Tsubasa look rather pale when compared to those offered by the latest GPUs (which are also a class of vector processors), such as NVIDIA’s A100, NEC believes that its vector processors can still be competitive, especially on datasets that require 48GB of onboard memory (as NVIDIA ‘only’ has 40GB).
As an added advantage, the NEC SX-Aurora Tsubasa card can run typical supercomputing workloads in a desktop workstation.
NEC does not publish prices of its SX-Aurora Tsubasa cards, but those who want to try the product can contact the company for quotes. In addition, it is possible to try the hardware in the cloud.
Sources: ITMedia, EnterpriseAI, NEC (via HPCwire)