How useful is Hyper-Threading for average user?

From the comments it comes down to testing with your application, if
you have a specific application which is dominant (eg, video processing).

Dbase apps are a cruel breed.
o Get as much in memory as possible
o Get memory bandwidth as high as possible
o Get the disk I/O subsystem on 64-bit 66Mhz+
o Burn the adaptec RAID card and stick quality in

Dual Xeon (re HT) is swamped by the /potential/ of Opteron re its
Hypertransport and similarly the Itanium2. That said, the HP 4-CPU
Itanium2 showed that good performance was possible with good Eng.
Market volume on such systems isn't gigantic re "importance" tho.

Re HT or non-HT, it comes back to application.
I'd prefer 4GB on my dbase server and non-HT than 2GB & HT.

A lot of s/w is badly written, but HT does offer promise.
The idea of multi-threaded components of a CPU die will grow:
o Eventually/presently the plan is 2-CPU per CPU-package
o One cools down whilst the other takes over re thermal load
o Vastly more advanced (and transparent) that HT

That said, it's not hard to envisage a world where it's not simply
"A" or "B" area of the chip package doing the work, but micro-mgt
of each 1mm^2 to linearise the thermal density over the whole area.

Not had to cool a 150W chip - 300cfm does 1500W, 30cfm does 150W.
Much much harder to cool small dense areas of silicon as output grows.
By 2010 I think it's 10kW per CPU, since die-shrinkage is underperforming
the growth in thermal output. Perhaps we have CPU the size of 8" floppies
:-)

So HT & likewise optimisation will probably become more important over time.
Which is perhaps what Intel & MSFT require to keep us all upgrading :-)
Why go A-E when you can make more profit going A-B-C-D-E.

Taxi services have been proving it for decades...
--
Dorothy Bradbury
http://homepage.ntlworld.com/dorothy...ry/panaflo.htm