The Dot Product Engine (Next Next)

[Curt_Hopkins]

The latest Next Next story is live. The Dot Product Engine (DPE), addresses the multiple-use accelerator architecture developed by Labs. The article explains how a team has created a  functioning, large-scale DPE demo, a Field Programmable Gate array (FPGA) with five billion transistors, a new data architecture to prepare digital problems for analogue computation, and multiple software layers. You can interact with an embedded version of the full-stack demo. 

One of the most common computational actions today is matrix multiplication, mathematical functions across columns of figures. One of the most popular expressions of this is object recognition and another is large-scale optimization and the location of global minima. But this action is very compute-intensive. Add to this the data flood from edge devices ranging from smart phones to sensors – which has made data analysis the key to finding signals in that noise, and artificial intelligence the key to data analysis – and you have a perfect storm. Too much data, too little information.

Hewlett Packard Enterprise’s Dot Product Engine (DPE) is a multiple-use accelerator architecture the company has produced to breast that storm. Most accelerators are designed to help solve specific problems. They’re one-offs, their creation is expensive, and they are irreplicable. With the DPE, you can spin up new accelerators with much less effort. And now, in the next step, an HPE team has created a full-stack DPE demo.

The team – consisting of Martin Foltin, master technologist; Sai Rahul Chalamalasetti, senior research engineer,;and Eun Sub Lee, design verification engineer; with assistance from many more across the company – has created a functioning, large-scale demo, a Field Programmable Gate array (FPGA) with five billion transistors, a new data architecture to prepare digital problems for analogue computation, and multiple software layers.

Read the full story here.