Case Study: Durham University accelerates cosmology research with Liqid composable infrastructure

Leading Cosmology Researchers Using Composability to Support Modelling of the Universe on COSMA8

Durham University’s Institute for Computational Cosmology, home to the COSMA8 national supercomputing facility within DiRAC, runs extremely large cosmological simulations that can take months and then require long periods of analysis. Because GPU and memory needs vary widely across workloads, the team needed a more flexible way to allocate expensive resources and improve utilisation. Durham University worked with Liqid to explore composable infrastructure for its HPC environment.

Using Liqid Matrix composable disaggregated infrastructure software, Durham University is composing NVIDIA A100 GPUs to servers so researchers can request only the GPUs they need, when they need them. The result is better GPU utilisation, more flexible acceleration for growing AI and simulation workloads, and a reduced carbon footprint by avoiding permanently populating every node with GPUs. Durham also sees future potential in composing memory across the fabric to support demanding simulations with lower latency and less congestion.

Durham University

Alastair Basden

Technical Manager