Author: Dann Corbit
Date: 17:13:04 12/22/99
Go up one level in this thread
On December 22, 1999 at 20:10:04, Greg Lindahl wrote: >But the SP interconnect hasn't been scaling in either bandwidth or latency -- >they now use fewer, higher bandwidth adapters, but the bandwidth per CPU is the >same as years ago, and fewer adapters mean you can't push many short messages >around. > >In comparison, the clusters that I build have more numerous, lower-bandwidth >adapters, with higher per-cpu bandwidth and higher short-message performance. I >also have 2 PCI busses per CPU. Is this the Beowulf model or something else? Is the clustering analogous to SMP or OpenVMS clustering or ??? Do you use message passing? How many CPU's can be built into the cluster? The newest COMPAQ machines can connect 32 CPU's with a bus speed of 1.6GHz. How does your system compare?
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.