Apols - Stephen, Peter. - PAX plan for phase 1b - track record - Outline for the work packages - What am I missing? - Updates - Peter D - Abouzied - what next for Parallel in time Update - Abouzied - realigning the structs to make more efficient for GPUs. Pointer chasing more problematic for GPUs. Small speedup shown, but much potential because the kernels themselves are very fast, on a low-ish-end GPU. Finding regular patterns should be part of the bid. Fusing these kernels - do we know which other DSL projects have been funded. What to do in phase 1B: 2 FTEs for 3 years (6 total). Start from engineering position and build up what is needed. Task engines - quick sched, work with openMP. But many things we need to include. Execute tasks on CPU and GPU. DSL idea of graphs. OneAPI - specify task graphs and have it generate. Task execution work package Abstraction work package - Matthieu surprisingly positive. Domain specific language - are there applications that work in real life? The fireDrake project with Pheonix would be an example. Write down a PDE, press a button and get a solution. Wonderfully flexible but may only solve a toy problem. Networking hardware. GPU v CPU balance Co-design Affecting standards (ieee etc). Communications - following on from Peter. Optimised data exchange. Code coupling project - need to leave the door open for this. Project titles. ExaPart, PAX, Part-Ex ExPeDiTe.