By: Michael Feldman

Cray has teamed up with Digital Catapult to help UK-based startups building artificial intelligence products get access to HPC resources in Cray’s Accel AI Lab.

Digital Catapult is a London-based company whose mission is to bring computer resources and expertise to UK startups specializing in advanced computer technologies. Currently, this encompasses artificial intelligence/machine learning, the internet of things, virtual and augmented reality, and blockchain applications. The company acts as a kind of agent, linking these fledging companies with other tech businesses and academic researchers to help them get their product to market.

For the AI space, Digital Catapult has developed something called the Machine Intelligence Garage, a program that gives startups access to HPC and other computational platforms supplied by third-party providers. The program also entails providing guidance on how to go about leveraging those resources and developing a commercial product. 

Through the program, every 6 to 12 weeks Digital Catapult holds an open call to find startups that it believes can benefit from computational resources of its partners. From the candidates that apply, approximately 30 startups are selected, each of which are provided a customized program for their particular product or service. The program, which is partially funded through the European Regional Development Fund and an Innovate UK grant, is scheduled to run for three years.

In this particular collaboration, Digital Catapult has partnered with Cray to get access to its Accel AI Lab, a cloud of sorts that offers a number of HPC systems built for machine and deep learning applications. It includes a CS-Storm 500NX system, each node of which has eight NVIDIA V100 GPUs; an XC system installed with the Urika-XC Analytics & AI Software Suite; a Urika-GX analytics system; a CS-500 system powered by Intel Skylake processors, and a ClusterStor storage system running Lustre.

“Our partnership with Cray opens a new door for UK machine intelligence businesses,” said Anat Elhalal, Head of Technology and AI/ML Lead Tech at Digital Catapult. “Cray’s expertise in high-performance computing will give startups much-needed resources to continue the amazing work already underway in the UK for AI and machine learning solutions.”

Despite the widespread use of GPUs and their availability for training neural network models, for cash-strapped AI startups, getting access to a system big and powerful enough to support commercial products remains a challenge. NVIDIA may have solved that problem to a certain degree with their latest 16-GPU DGX-2, but the $399,000 price tag limits its market to customers with relatively deep pockets

“Based on research conducted by the Machine Intelligence Garage, startups in the UK have identified that one of their biggest obstacles to progress in AI is access to exceptionally powerful and fast computational systems,” said Per Nyberg, vice president of market development, artificial intelligence and cloud at Cray. “We’re seeing a convergence between AI and supercomputing as organizations work toward taking AI projects out of beta and into production. The number of AI models that must be trained will only increase and organizations are coming to realize the performance and I/O challenges that will arise.”

Nyberg also noted that research conducted by the Machine Intelligence Garage found that trustworthy benchmarking and performance modeling will be increasingly important as the machine learning ecosystem evolves. “This is an area where Cray sees a similar need and is also echoed by its customers,” Nyberg told TOP500 News.

Cray’s AI Accel Lab was announced in November 2017, but this is the first instance where its use by potential customers was made public. According to Nyberg though, there have been a number of engagements going on behind the scenes. “We’ve had both some prospective and existing customers use the Cray Accel AI Lab and demand has been steadily increasing since it came online,” Nyberg said.

Besides Cray, Digital Catapult also partners Google, AWS, Graphcore, SpiNNaker, EPCC, the Hartree Centre, Newcastle University, and NVIDIA to help support the Machine Intelligence Garage.