Re: hyperthreading or not
On Tue, Jan 24, 2006 at 06:02:48PM -0500, Adam Skutt wrote:
> Lennart Sorensen wrote:
> >They aren't intel netburst architecture. Someone else is free to
> >implement hyperthreading (SMT) without a shared pipeline after all. Not
> >sure how Sun (or IBM for that matter) do theirs.
> They have multiple cores per die with multiple SMT units per core. And
> it's very effective for J2EE websites and similar: short, mostly I/O
> bound code that doesn't take a lot of time.
>
> The point is that generalizations such as, 'A vendor is incompetent if
> they don't ship HT off' is just retarded because even on a Xeon, HT can
> have noticeable benefits for /some/ serving tasks. Not all, but some.
Absolutely. Of course some people would be tempted to think any vendor
shipping xeons instead of opterons is making a bigger mistake than
someone shipping with HT enabled. :) It all depends on what your work
load is and what stupid politics you have to put up with. On the other
hand it will be interesting to see how things go when intel finally gets
new servers out based on the new Pentium-M/3/2/Pro derived core.
Len Sorensen
Reply to: