Agenda
8.30 Introduction
8.35 - 9.20 Keynote : Adrian Jackson, EPCC – Introduction to a new AI inference benchmark framework https://www.teasbench.com/, analysing performance, cost and energy of various accelerators under different workloads
9.20-9.40 : EcoShift: Performance-Aware Power Management for Power-Constrained Heterogeneous Systems
9.40-10.00 : From GPU-Accelerated Systems to Modern NVIDIA and AMD Superchips: Analyzing Energy and Power of a Heterogeneous SYCL Application
10.00-10.30 Coffee break
10.30-10.50 : Compositional Energy Modeling of Layer-Resolved GPU Training
10.50-11.10 : The Economic Value of Heat Reuse
11.10-11.30 : Fast and Accurate ML Surrogates from Simulation: An HPC Liquid-Cooling Models Case Study
11.30-11.50 : Per-Job GPU Frequency Selection for Production Density Functional Theory from Submission Inputs
11.50-12.00 : Does New Hardware Improve Carbon Impact of HPC systems?