The most popular and comprehensive Open Source ECM platform
Let us know more
AI Chips: Moving Beyond the GPU
GPU – Graphics Processor Units. GPU chips power most of today’s AI technology. These chips have been very successful for processing certain types of AI algorithms, but as the name GPU implies, GPUs were designed to speed up graphics processing, not for AI processing.
In the next few years expect to see a move away from the close connection of AI technology and GPUs.
Nigel Toon, CEO of Graphcore, said that “all the innovators we spoke to said [using GPUs] is holding them back from new innovations. If you look at the types of models that people are working on, they are primarily working on forms of convolutional neural networks because recurrent neural networks and other kinds of structures, [such as] reinforcement learning, don’t map well to GPUs. Areas of research are being held back because there isn’t a good enough hardware platform, and that’s why we’re trying to bring [IPUs] to market.”
The need for more targeted AI chips has led to a recent flurry of activity by almost all the major tech firms to create a new generation of chips that are designed specifically for AI processing. The companies include Amazon, Apple, Facebook, Google, and Alibaba.
A Deloitte whitepaper said that “these new kinds of chips should increase dramatically the use of machine learning, enabling applications to consume less power and at the same time become more responsive, flexible and capable.”
