TechSocial Series - June 2026

Abstract

At Callosum we are working on how heterogeneous computing systems made up of multitudes of different accelerator types can be used to run large scale AI more efficiently than when using commonly built homogeneous computing systems. This presentation will cover the results we have from our existing multi-cloud deployment software system that achieves the processor heterogeneity by orchestrating workloads across multiple clouds by identifying and deploying novel disaggregation patterns for large scale AI. As the latency of multi-cloud deployments naturally can be prohibitive for advanced heterogeneous deployment, we will also cover the heterogeneous cluster we are in the process of constructing, which will allow for even more complex and optimised deployment strategies. With this presentation we hope to start an active collaboration with the Advanced Research Computing Centre on heterogeneous computing systems for AI and scientific workloads deployed at UCL.

Free regsitration available via Pretix

Head to our TechSocial Series hompage for more information about future dates and speakers

Map