Microsoft to Integrate Graphcore's AI Accelerators into Azure Cloud Platform


Users of the Microsoft Azure cloud platform will soon be able to count on an acceleration of calculations related to artificial intelligence provided by the specialized chips of the English start-up Graphcore. The announcement of this partnership was made Wednesday, November 13, 2019. Microsoft had been part of the investors of the $ 200 million fundraising carried out by Graphcore last December.

A first hyperscale client

This is a major victory for the British start-up founded in 2016. Graphcore says in a statement that this proves that its technology is mature, whether hardware or software architecture. The trust of a "hyperscale" actor (a term describing cloud giants, whose server infrastructures are incomparable to those of traditional companies) is indeed a guarantee of quality, in addition to representing a huge commercial potential. Nigel Toon, CEO of Graphcore, has welcomed the fact that Graphcore technology will soon be available around the world.

Specialized chips as an alternative to GPUs

Graphcore develops chips that it named Intelligence Processing Units (IPU) to differentiate them from existing CPUs and GPUs. Their architecture is highly parallel, like that of GPUs, but is optimized for calculations related to artificial intelligence, and in particular the processing of natural language. The start-up says that training models for image recognition, voice processing or video analysis is 10 to 100 times faster using its products than with a conventional system. It further specifies that its architecture is equally adapted to the inference.

The areas of application are vast: health, industry, finance … The French Qwant uses it for its image search engine, and reports performance 3.5 times faster than before. Carmot Capital, a California firm specializing in risk hedging, reports calculations that are 26 times faster. Graphcore currently employs about 200 people and has projected $ 50 million in revenue for 2019 early in the year. It plans to reach $ 1 billion in revenue within 5 years.

The fight with Nvidia and Intel will be rough

Obviously, the competition is fierce. Intel, which remains the king of datacenters with its CPU, has just presented its chips Nervana, direct competitors of the products of Graphcore. Nvidia continues to improve its GPU, which remains today a preferred solution for deep learning. We can also mention Google, which has developed its own in-house specialty chips, or Cerebras, a rival California start-up.


We will be happy to hear your thoughts

Leave a reply

AB Smart Health
Enable registration in settings - general
Compare items
  • Total (0)