
as a former alumnus of EPCC
GO TEAM EPCC!
Hopefully, as the author stated rightly, this would encourage other UK Universities with expertise in HPC to join in the challenge such as Imperial, Bristol, Cambridge and Southampton to name a few.
9 publicly visible posts • joined 3 Jul 2010
I'll be there in Leipzig this summer. It's going to be tough to beat the Chinese teams even though I am rooting for EPCC (University of Edinburgh).
I wasn't really aware that you are allowed to also hand tune the codes yourself as it is not clearly specified in the regulations of the HPC Advisory Council. If that's the case, then this competition is actually more exciting than I previously thought.
I am studying the MSc. in HPC @ Edinburgh Univ. and distinctively remember a lecture on Exascale computing held by a Professor heavily involved in the HPC area within the UK and EU regarding the need for more and more computing power especially for weather forecasting. This is because as some have already described above, in order to have better medium and long term predictions, you need to compute bigger and bigger areas. With a normal distributed system, if you decompose millions of these sub areas across machines scattered around the world, the intensive halo swapping (neighbourhood borders) communications would absolutely grind the entire computation to a halt. This is why you need more cores and faster interconnects for being able to throw a bigger problem at the beast (Gustafson's Law) which in turn should hopefully mean a more accurate prediction in a faster manner.
I wouldn't necessarily blame their models to be wrong, just their capabilities on how much data they can actually use them on which for this particular science seems to be a rather important point to take into account.