Slashdot Mirror
£52 Million Govt Funding for New UK Supercomputer
Lancey writes "The BBC reports that the UK government has contributed £52 million towards the building of the High-End Computing Terascale Resource to replace two existing supercomputers currently in use by British scientists. The story claims a maximum speed of 100 teraflops, although it is unlikely that the machine will ever be pushed to this limit. Some of the government funding will also be used to train scientists and programmers to develop software capable of exploiting the machine's potential."
It's not a world record (IBM's Blue Gene/L is pushing 280 TFLOPS), but it's still very fast. It will almost certainly be in the Top 10 by the time it comes online.
In short, they're not. That's the BBC's attempt at explaining why the theoretical peak isn't practically achievable. See http://en.wikipedia.org/wiki/Teraflop#FLOPS_as_a_m easure_of_performance
Iain.
Of course it's 100,000 times faster than an ordinary computer. It's a rack of 100,000 ordinary computers.
Anyone remember the days when the word 'supercomputer' actually meant something?
Yes! and good riddens.
Do you remember having to re-code for every single machine? Because they were such specialized machines, they tended to be extremely fickle: one wrong operation and performance would go down the drain.
In practice, most computational work in the end consists of running many jobs independently. There are rare occasions where a single super fast CPU might be better but it's even rarer for the performance gains to outweigh the incredible cost increases for buying specialized supercomputer hardware.
Whether it's wise to spend so much money on a single enormous cluster is another issue. You could buy many many individual clusters for individual groups and have them operational in a matter of weeks, rather than having wait till 2008. Besides, the thing is going to be obsolete by 2010.