Graphcore's CEO Nigel Toon has said that access will be prioritized for those focused on pushing the boundaries of natural language processing (NLP) and machine intelligence. Toon said this is the first time a major public cloud vendor has offered Graphcore IPUs (Intelligence Processing Units) and therefore represents a breakthrough for the startup.
BMW, Microsoft and Dell are all strategic investors in Graphcore and helped the company raise $200 million about a year ago, taking the amount invested in Graphcore to $300 million and valuing the company at $1.7 billion.
Graphcore has been working for more than two years with a team at Microsoft, led by Distinguished Engineer Marc Tremblay, which has been developing machine vision and NLP systems for Azure.
In a blog on Graphcore's website Toon also said that Graphcore will be demonstrating its IPU technology with Dell at the upcoming Supercomputing Conference (International Conference for High Performance Computing, Networking, Storage, and Analysis) in Colorado.The Dell EMC DSS 8440 can deliver more than 1.6 petaflops of machine intelligence compute, using 8 Graphcore C2 IPU-Processor PCIe cards, each with two IPU processors on board in a 4U-sized chassis.
To support networks of IPUs Graphcore has developed the Poplar software stack, which integrates with TensorFlow and Open Neural Network Exchange (ONNX) allowing developers to use their existing machine intelligence development tools and existing machine learning models. Initial PyTorch support is available in 4Q19 with full support becoming available in early 2020.
Graphcore has also published a selection of performance benchmarks based on the IPU systems that are now available as an IPU Cloud preview on Microsoft Azure and as IPU-Server products from Dell. These mainly compare IPU networks with 150W TDP against GPU solutions at twice the maximum power consumption.
Related links and articles: