Nikolai Ardey - Volkswagen

Excecutive Director - Volkswagen

Future vehicle compute platform - approaches to meet the challenge
Artificial Intelligence (AI) applications are on a massive advance on towards future vehicle compute platforms. Machine Learning (ML) models are rapidly mutating, getting bigger and more complicated pushing even dedicated and highly specialized hardware (HW) to the limits and beyond. Today, we have three basic options: CPUs, GPUs, and ASICs.  CPUs, designed for general purpose compute, are flexible, but not powerful enough for AI.  GPUs, designed to render triangles, are powerful but not efficient enough for AI.  ASICs, by definition, can be hardcoded for specific ML models making them both powerful and efficient enough for AI. However, they may not be flexible enough to adapt to future AI approaches, which are hard to be anticipated up to the needs ofdue to long HW development time scales. Hence, future AI needs HW that is the best of all three:  as flexible as a CPU, as powerful as a GPU, and as efficient as an ASIC. A generic, programmable HW-dataflow design, drawing compute efficiency from the typical emerging ML matrix structures that are sparse having with lots of zero’s and near zero’s and a SW architecture capable of utilizing those benefits are the major design trends heading into that direction. This keynote is to exploratewill explore the potentials of these approaches.