On Tue, 2002-12-24 at 01:58, Matthew Garrett wrote: > In chiark.mail.debian.devel, you wrote: > >On Sat, 2002-12-21 at 19:36, Adam Majer wrote: > >> Damn it! What about a P4 distribution???!!? > >> > >P4s are i686s. > > Commonality of instruction set does not necessarily imply commonality of > optimisations. The PIV is quite different from a PPro, > architecture-wise. A P4 distribution would probably be a win over a 686 > distribution - on the other hand, the jump probably wouldn't be as big > as from 386 to 686, which nobody seems to have demonstrated as being > terribly necessary anyway. > Aside from truly processor-centric tasks such as crypto, I've not seen any indication that there's a win from processor-specific optimisation. Most software is far more likely (I'd think, anyway) to get bogged down in such arch-specific bottle necks as IO. A quick google brings up lots of papers about the performance differences between different C compilers, but nothing about the benefits of compiling with different processor options using the same version as gcc. Has anyone got some real-world evidence that this makes a difference? I'm thinking about things like how many pages/second an Apache server + mod_perl can serve when compiled generically over when compiled with the "perfect" optimisations for the processor. Scott -- Scott James Remnant Have you ever, ever felt like this? Had strange http://netsplit.com/ things happen? Are you going round the twist?
Attachment:
signature.asc
Description: This is a digitally signed message part