Re: kernel 2.6.18+ SMP question
On Sat, 2007-10-13 at 15:21 -0700, Andrew Sackville-West wrote:
> On Sat, Oct 13, 2007 at 09:30:57PM +0200, Robert Cates wrote:
> > michael wrote:
> >> On Fri, 2007-10-12 at 22:02 +0200, Robert Cates wrote:
> >>
> >>> Hi all,
> >>>
> >>> just a general question - I'm interested in getting a motherboard that
> >>> supports 4 of the new AMD Quad Core CPUs. That would be effectively 16
> >>> CPUs. My question is, with the 2.6.18-5-686 kernel from etch, will the
> >>> SMP kernel work with all 16 cores? What is actually the current limit of
> >>> cores (CPUs) that the 2.6.18 (or newer) kernel will support?
> >>>
> >>
> >> don't expect things to go 16x faster
> >>
> >>
> >>
> > Ok, but 16x better? Or 16x more efficient? I know that if we're talking
> > about a 2GHz quad core CPU we're not getting 8GHz of speed, but what
> > exactly is the (performance) advantage of SMP?
> >
>
> how about 16x more parallel? at least to the extent that your workload
> is able to parallelize (is that a word?). IOW, if you have lots of
> tasks running independently of each other and/or you have tasks
> running code that can take advantage of parallel processing, then
> those things that fit that criterion will run in parallel. And those
> tasks will then complete faster because they have more cpu time than
> they would get in a system with fewer cpus.
>
> At least that's how it seems to me.
current multicore also shares (L2?) cache so there's contention there
too which will severely affect codes that require data to work on...
Reply to: