The traditional argument goes like this: an n-processor machine can never solve a problem more than n times faster than the same machine with only one processor
This is not quite true: it's not unknown to see superlinear speedup in some "embarrassingly parallel" calculations, because, for example, a larger proportion of the dataset will fit into cache, so that you spend less time waiting for memory fetches.
![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png)
timeasmymeasure
Attack of the pedants
Date: 2005-09-21 01:43 pm (UTC)never solve a problem more than n times faster than the same machine
with only one processor
This is not quite true: it's not unknown to see superlinear speedup in some "embarrassingly parallel" calculations, because, for example, a larger proportion of the dataset will fit into cache, so that you spend less time waiting for memory fetches.