To hyper-thread or not to hyper-thread

[Aaron Konstam]

On Mon, 2009-12-07 at 22:16 +0200, Jussi Lehtola wrote: 
> On Mon, 2009-12-07 at 09:44 -0700, Greg Woods wrote:
> > On Mon, 2009-12-07 at 16:18 +0100, Joerg Bergmann wrote:
> > > The problem of the Pentium 4 D: It is not really a dual core one.
> > > Hyper-Threading means: There is one core with two execution paths, which
> > > means some of the common CPU features, but not all, are present twice.
> > 
> > One feature in particular that is not present twice is some of the
> > caching. This is sort of why they named it "hyperthreading". If you can
> > get multiple threads of the same process, sharing the same memory, to
> > run simultaneously, there is a performance boost. But if you try to run
> > two completely different processes simultaneously, there will actually
> > be a performance LOSS because of all the cache misses this will cause.
> 
> This may not be true - in the high performance computing community
> hyperthreading is usually not used, since if you're cpu bound, then
> execution is about 20% faster in without hyperthreading since no
> performance is lost because of the dual core emulation.
> 
> However, in normal desktop use you don't really care about the MFLOPS;
> hyperthreading makes the system more responsive.
Well the Intel ads make XEON hyper-threading sound like the greatest
thing since sliced bread, 
> 
--
=======================================================================
It would be illogical to kill without reason. -- Spock, "Journey to
Babel", stardate 3842.4
=======================================================================
Aaron Konstam telephone: (210) 656-0355 e-mail: akonstam at sbcglobal.net