Alastair, Tobias, Stephen, Richard, Matthieu, David, George, Peter, Ben, Tom, Scott, Edo. - Actions Action: email around working group meetings that are known about. Action: Richard to discuss with Andy Turner about microbenchmarks. Action: Richard to chase Katie about response to ARM. Action: Ali - investigate extreme scaling workshop on dirac facilities, similar to german systems a few years ago. Action: Richard - talk to Jeremy about how to proceed for phase 1a. Action: Ben - contact Sylvain. Action: Richard - talk to Scott about possibilities of linking up. Action: Richard - talk to Simon about whether interested.   - General updates CiSE article sent off (Richard, Matthieu, Ben). - Phase 1b application (Richard) Hoping to work with molecular dynamics group - they are mostly talking about available codes, but don't develop software - this is done separately, possibly commercially. More interested in standards and APIs. Tobias and Richard talked to AEA to convince about task-based programming. But they weren't interested. This would rock the boat and mess things up. Already moving forward from F77 to F90 - this was enough of an advance. Need alternative to go into phase 1B and see what they say. Richard will talk to Jeremy for some advice. Could try linking with turbulence people - would they be interested in task-based parallelism? Also MHD that might be of interest to them. Do they already have much investment in development of other codes that don't want to consider task-based? Ben to contact Sylvain. Simon Macintosh-Smith may be worth investigating, but we don't want to get into DSLs. But ultimate way of doing separation of concerns. Writing a layer under the DSL with an API. - Proposals funding update - Tobias, task parallelism (successful) - Dave - time parallelism) (successful, KO meeting has started) - Code coupling (Ben/Stephen led by Garth Wells) (requested responses to reviewer, a bit haphazard. One 9.5, one 8, one black, non-technical) - Graphics (Dave) (pending, project description written, waiting for a student to choose this project.) - Future paradigms (RGB/DiRAC) (pending) - Quantum computing (RGB) (pending, interview tomorrow) - Contracts (Richard)   - Hydro weak-scaling (Edo) Repeating scaling tests on cosma8, looking at 2,4,8 ranks per node, 256^3 particles per tile. Scaling 4 magnitudes in number of cores, still seeing very good performance. Will look at how 2,4,8 compare with each other. Running 1.4x faster per core than cosma7. Can get 1Bn particles per node, much more than cosma7.   - RDMA updates (Peter, not this week) Nothing to add about the RDMA. Parts of development decoupling memcopies from send/receive being put into the main code, and reducing the message sizes. Can now trace local copies from buffer to buffer, so clearer picture of where slowdown comes from. Numa awareness. Ideal 1 rank per node to minimise overheads. Doesn't do will with numa regions. Some work on progressing this. Ideal would be to launch a task on the core where the data is, but hard. Good speedup in test runs but not so good in practise. - GPU update (Abouzied) Trying to sort out cuda code to launch within swift - struggle to sort out compiling using autotools with cuda. Set up some code to send/recv from GPU, so this blockage is now sorted out. Will start working on setting up some sph density code and compare with swift. A lot of this was very cuda specific - so might be hard to port to opencl. Will do this later - test out what we have now. - Individual workpackage update [all] - Training (Tobias) Next monthly workshop on Thursday. German computing centre organised extreme scaling workshop many years ago - 1 week, get whole system for test runs. Is there some room to organise a workshop - select number of groups who can run on 5k cores or more, access to system, get a few mpi people over, help with tuning mpi, run over night. Could be benchmarking group, could be hardware group, etc. - Outreach (?)   - Others/AOB? PhD funding from Dell - to work on AMD technologies. Next meeting: 5th July.