Platforms for AI

Accelerating science with a dedicated platform for AI workflows

AI-driven research initiatives at the Sanger have expanded in scope significantly over the past few years, with new use cases for the emerging technology across a wide range of research initiatives.

The applicability of AI tooling is set to explode with the increasing volume of genomic data from the rapid expansion of sequencing technologies across the world.

Our HPC resources are challenged to exceed the requirements of AI workflows on the biological data of today and tomorrow. We host 275+ GPU in configurations described on the farm itinerary page for common use, and have additional dedicated accelerator resources for specific research groups whose work is heavily AI-dependent.

The GPU are linked with ultra-fast 400Gb/s networking, allowing fast and reliable transfer of data to and from on- and off-site storage platforms.

Multi-Node Support

The hardware used to support AI workflows on the farm is arranged into multiple nodes, each containing multiple GPU (including 2x, 4x, 6x, and 8x nodes). For significant workflows whose data processing requires multiple nodes, we support running across multiple nodes to leverage the acceleration capabilities of up to 16 GPU simultaneously.

MLFlow Platform

We host a dedicated web-based platform for machine learning operations based on MLFlow, developed in-house by our team of RSE.

This system offers a lightweight interface to machine learning and AI workflows for users who are starting out, and provides much of the necessary groundwork to maintain reproducible and repeatable AI pipelines, which benefits all research at the Sanger.

Programme Support

To support research programmes in their adoption of AI we offer long-term consultation and cooperation with research software engineers and research technology specialists (see our page on RSE, RTP resources).

Research software engineers have allowed teams of researchers to fully leverage the benefits of an AI-ready cluster, by developing standardised tools for submitting multi-node jobs and cooperating with teams to adapt their workflows.