Fear of Thinking War Machines May Push U.S. To Exascale 192
dcblogs writes "Unlike China and Europe, the U.S. has yet to adopt and fund an exascale development program, and concerns about what that means to U.S. security are growing darker and more dire. If the U.S. falls behind in HPC, the consequences will be 'in a word, devastating,' Selmer Bringsford, chair of the Department. of Cognitive Science at Rensselaer Polytechnic Institute, said at a U.S. House forum this week. 'If we were to lose our capacity to build preeminently smart machines, that would be a very dark situation, because machines can serve as weapons.' The House is about to get a bill requiring the Dept. of Energy to establish an exascale program. But the expected funding level, about $200 million annually, 'is better than nothing, but compared to China and Europe it's at least 10 times too low,' said Earl Joseph, an HPC analyst at IDC. David McQueeney, vice president of IBM research, told lawmakers that HPC systems now have the ability to not only deal with large data sets but 'to draw insights out of them.' The new generation of machines are being programmed to understand what the data sources are telling them, he said."
Re:wow, stupider than MAD! (Score:5, Informative)
Re:wow, stupider than MAD! (Score:2, Informative)
High-performance computing = HPC
exascale = scale on which supercomputers are capable of 1 or more exaflops
1 exaflops = 1000 petaflops = 1 quintillion floating point operations per second (FLOPS) = 10^18 FLOPS
Re:Fund us or [insert fud] (Score:4, Informative)
The big new Chinese supercomputer is ~ 34 petaflops. Exascale is 1,000,000 petaflops. That is a pretty big difference in scale. Although current supercomputers have tended to be "more of the same" stacked higher, the difference in scale here may signify a difference in kind. At the least there are likely to be some new engineering and programming challenges involved if they are really going to exploit that kind of potential.
Re:Fund us or [insert fud] (Score:5, Informative)
1 exaflops (10^{18} flops) is 1000 petaflops (10^{15} flops). The Chinese are 3.4% of the way there. Exaflop-scale computers are realistically expected for 2019, i.e. in 6 years' time.
Meanwhile, India is supposedly building a 140 exaflop computer for 2017 [defencenews.in]
Better get a move on I guess.