On 08/08/2012 09:54, David Matthews wrote:
Actually, one has to be careful when interpreting the parallelism figures. Much of the time an ML application is memory-bound which means that using more processors can make an improvement in overall run-time but the overall CPU usage will grow. As far as I can tell, the time that a processor spends waiting for memory counts as its CPU time unlike when it is waiting for IO.
Just in case that wasn't clear from the context, by "memory-bound" I meant limited by the bandwidth between main memory and the processors.
David _______________________________________________ isabelle-dev mailing list isabelle-...@in.tum.de https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev
