Intel to release AI chip line to compete with Nvidia‘s GPUs

New line of chips will improve deep learning training times.
(FinancialPress) — Intel may be a late-comer to the AI party – but after this announcement, no one can say that they‘re not arriving in style. The company is setting itself up to stomp on their market grounds with authority by announcing the release of a family of chips dedicatedly oriented towards artificial intelligence. The new product line will be called the Intel Nervana Neural Network Processor family – or NNP.
The main orientation for the NNP family is to cover the demanding needs of machine learning. It will be geared not towards PCs, but rather to data centers. While server stacks are still predominantly formed by Intel CPUs (it‘s estimated that they enjoy a 96% share in the data center market), AI workloads are better carried by graphical processors (GPUs) developed by firms like ARM or Nvidia. Therefore, these companies are enjoying a steady and very accentuated growth in sales lately – with Nvidia gaining a healthy 56% rise in year-on-year revenue this far into 2017.
Already in the fray is Google, which has designed its own silicon line under the name of Tensor Processing Unit (TPU). It is currently used to cover the needs of their cloud computing business. Also worthy of note is UK‘s Graphcore, which has also thrown its hat into the arena.
Instead of starting from scratch, Intel has reached into its deep pockets and cherrypicked the best talent in the AI market. The purchases began with deep learning startup Nervana System in Augusto 2016; then came vision specialist Mobileye back in March; then came the acquisition of Movidius in September. After completing their team, they began teasing their Neural Network Processors under the codename “Lake Crest“. Nervana was the main influencer in the development of the NNP chips and Intel has said its used the company‘s expertise to achieve faster training time for deep learning models
What's your reaction?
Happy0
Lol0
Wow0
Wtf0
Sad0
Angry0
Rip0
Leave a Comment