Bloomberg, and others, report (£) that Sunak is planning to spend £800m on a new British Supercomputer. This is also covered in the Register, who among other things suggest that the amount of money is extraordinary for a super computer, allowing the building of the largest in the world, exceeding the computational power of the US Frontier system, which is over three times faster than No 2, and nearly four times faster than the Finnish No 3, the LUMI project. The UK currently has none of the top 10 , its best is 28th, and it withdrew from the EU’s HPC joint undertaking because of Brexit.
In this article on linkedin, or on Medium, I explained that the advantage of co-operation with other countries/projects was leverage of the investment. HPC grids can co-operate, even over wide areas, depending upon the basic algorithmic pattern being used for the specific solution, the task synchronisation costs and the number and duration of compute threads required. It seems to me that this is another good reason for increasing the co-operation with EU either through Horizon or specifically re-joining the HPC joint undertaking.
The Register suggests that the UK project if it gets funded will also ensure that the chips come from the UK; it surprised me to learn that there are 11 chip fabs in the UK, two of which are known to be capable of over 20,000 chips/month but none are in the top 100. As HPC implementations move from intel towards ARM, local sourcing becomes more feasible, but the industry could obviously do with the help. However last time I looked, the critical design point was the CPU interconnect which was generally designed for the purpose, although from examining the Top 500 report, it seems that Gigabit Ethernet is the most popular interconnect.
Originally published at https://www.linkedin.com.
