On Sat, Feb 17, 2007 at 12:16:58AM +0100, pinniped wrote: > > 'Multicore' processors are pretty new - older machines have always had an > enormous number of processors. 80 cores on a single chip would take up most > of the silicon wafer. (and will probably have a few defective cores and > rather interesting cooling problems) > Are you sure? When I took a semiconductors class (about four or five years ago), the professor was talking about how the state of the art in fabrication was being able to pull 18 inch-diameter silicon slugs. I would imagine that things have improved. Not just that, but we are now seeing the bug manufacturers getting to 45 nm feature sizes. You can fit *lots* of stuff on a wafer that big at 45 nm. But, I don't think 80 cores would take all or even a large part of a wafer. Granted, the cooling issues would be interesting. > > for Question #2: > Microsoft will seriously struggle - there are numerous defects in the > system and we always have a good laugh when Microsoft talk about > reliability, 'high performance', and 'scalability'. WinDos cannot even > handle time properly; in contrast, UNIX was handling time even before DOS > was out on the market (allowing every user to have their own time zone - in > fact any program running can have its own time zone). > This won't stop that marketriods from convincing lots people that windows is actually able to handle that many cores. I always chuckle when I see a mention of Windows Server Cluster Edition and think of the poor saps who get stuck administering that. > Apple's OSX is based on BSD, which will currently run on at least '8-way' > machines. I don't keep in touch with BSD developments so I don't know what > their current limit is. > Mac OS X uses the Mach-based XNU kernel. The BSD parts that are used are the userland. > > Of course more cores doesn't mean a faster machine unless you do things in > parallel. It may be great for servers which do a lot of actual processing > and in astronomy for 'multibody' calculations (and in physics for 'Monte > Carlo' simulations of light scattering) but for most purposes it will just > keep the room that little bit warmer. > Don't forget CFD! Regards, -Roberto -- Roberto C. Sanchez http://people.connexer.com/~roberto http://www.connexer.com
Attachment:
signature.asc
Description: Digital signature