Sign In

Communications of the ACM

ACM TechNews

Are Supercomputers Just Better Liars?

View as: Print Mobile App Share:

Supercomputers' superior power compared to ordinary machines does not make their predictions unerring, argues consultant Andrew Jones. He observes, for instance, that computer simulations are increasingly substituting for physical testing for most design work, which saves time and money. Other areas where computer predictions are increasingly relied on include climate forecasting and business decisions, which makes the issue of the computer's accuracy all the more critical.

Jones says that supercomputers are frequently used in science and engineering to check the validity of computer models, but model users often fall into the trap of assuming that the prediction is more accurate because of the supercomputer's higher resolution. Jones counters that higher resolutions are no guarantee of accurate results, and he notes that "at the higher end of supercomputing, fault tolerance is becoming critical — not just in node failures, but in softer errors such as data corruption in memory or in the interconnect."

Improved modeling will be facilitated by the provision of additional computing power via high-performance computing, but Jones cautions that some of that computational power should be committed to the validation of the modeling. He also advises against making "the dangerous assumption that physical testing is always better than computer predictions. Physical testing has its own sources of errors, assumptions, and regions of validity."

From ZDNet UK
View Full Article


No entries found