W
e tested the new dual-core
Pentiums using a Pentium
Extreme Edition 840 on an
Intel 955X motherboard
equipped with an ATI Radeon X850 Pro,
1GB of Crucial Tech DDR2/667, and a
160GB SATA Seagate hard drive with
NCQ enabled. Because the Pentium D is
nothing more than a Pentium Extreme
Edition with the Hyper-Threading abili-
ties turned off permanently, we simulat-
ed Pentium D performance by disabling
Hyper-Threading in the BIOS. When we
finished testing dual core, we installed a
3.73GHz Pentium 4 Extreme Edition with
2MB of L2 cache in the same mobo and
reran the tests with Hyper-Threading
enabled.
To test dual core, we ran both multi-
threaded and single-threaded apps to
simulate a real-world experience. Some
of the tests, such as Doom 3 , Quake III ,
Photoshop CS , and 3DMark2005 , you’re
already familiar with. We also ran some
newer multi-threaded applications such
as Abbyy FineReader 7.0 Pro. This OCR
(optical character recognition) app was
recently revised to support multi-core
systems. We took a scanned, 133-page
excerpt of the Warren Commission
Report in PDF format and ordered the ap-
plication to OCR it, then timed the result.
DVD Shrink is a popular multi-threaded
application used to compress DVD-9
content to fit on a single-layer recordable
DVD. To keep the extraction speed of
an optical drive out of the equation, we
copied the contents of the Terminator 2
DVD to our hard drive, then tasked DVD
Shrink with recompressing the disc’s
contents to fit on a single 4.7GB disc.
For additional multi-threading tests, we
grabbed Newtek’s LightWave 7.5D and
used it to render a single frame of the
built-in benchmark raytrace file using
eight separate threads.
Because few of today’s applications
are multi-threaded, we also threw sev-
eral multitasking tests at the dual-core
system by simply running two bench-
marks simultaneously, then reported the
scores for each benchmark separately.THE RESULTS
Scan the benchmark chart, and the first
thing you’ll notice is how much fasterthe dual-core CPUs are in multi-threaded
applications when compared with the
3.73 P4 Extreme Edition. The latter’s
533MHz clock-speed gap, 2MB of L2, and
1066MHz front-side bus just can’t make
up for an additional core when you run
compute-intensive tasks. In our multi-
threaded OCR test, for example, the PEE
is 69 percent faster than the P4EE. The
PEE also transcoded our DVD 33 percent
faster than the P4EE with HT. The PEE
completed the LightWave benchmark 60
percent faster than its higher-clocked
sibling. The PEE blew away the P4EE in
the majority of our homegrown multi-
tasking tests.
As expected, the PEE doesn’t fare well
with non multi-threaded benchmarks
such as MusicMatch 10 and Photoshop
CS. Multiple cores just can’t compete
with raw horsepower in single-threaded
apps. We saw similar results in the gam-
ing tests. This is why Intel isn’t pushing
the dual-core Pentium Extreme Edition
or Pentium D as gaming procs. The
superior bandwidth (from the larger L2
cache and 1066MHz front-side bus) and
higher clock speeds of the Pentium 4
own gaming.
Oddly, we found that the performance
of the PEE improved in some tests when
Hyper-Threading was turned off. Why?
Intel naturally blamed Microsoft, postu-
lating that the scheduler in Windows XP
doesn’t differentiate between Hyper-
Threaded virtual procs on the same
core, and as a result, it throws all the
work to one core only, instead of evenlyspreading work between the two physi-
cal cores.
Our tests that mixed multi-threaded
applications with single-threaded apps
seem to support this theory. In such sce-
narios, the multi-threaded applications
had enough work to keep both physical
cores busy and generally the PEE with
HT enabled did better than the same CPU
with the HT disabled (in other words, the
Pentium D).THE VERDICT
Overall, we’re impressed by the perfor-
mance benefits of dual core, but only
with the right workload. Multi-threaded
applications and multitaskers should see
massive boosts in performance over sin-
gle-cored processors. Things you could
never do before, such as ripping a DVD
while playing an online game, are quite
possible with a dual-core processor.
Gamers who demand the fastest frame
rates and users of primarily non multi-
threaded applications, however, should
stick with the fastest single-core proces-
sor available. That said, if you’re willing
to sacrifice a few fps in games, you can
pick up a huge performance boost when
you use multi-threaded apps. And with
modern games relying heavily on the
GPU, you don’t give up too many frames
by opting for a dual core. Of course we
won’t know the full story until AMD
releases its X2 series of dual-core procs
in a month or two. Only then can we
declare the real winner. ■CPU PENTIUM EE HT OFF PENTIUM EE HT ON PENTIUM 4 EE HT ON
Clock/FSB 3.2GHz/800 3.2GHz/800 3.73GHz/1066
3DMark03 overall 12,335 12,321 13,020
Doom III (fps) 86.5 85.0 97.4
Quake III (fps) 384.1 384.5 484.5
3DMark05 5,731 5,729 5,867
Abbyy FineReader 7 Pro (sec.) 210 161 272
DVD Shrink 3.2 (sec.) 477 443 587
LightWave 7.5D Raytrace (sec.) 61.2 50.4 80.1
MusicMatch 10 MPC 2004 (sec.)^276280^238
Photoshop CS MPC Test (sec.) 363 356 324
3DMark05 CPU w/ DVD Shrink 5,677 4,369 5,175
DVD Shrink w/ 3DMark05 CPU (sec.) 767 573 879
3DMark2005: w/ MusicMatch 10 4,528 4,457 3,399
MusicMatch 10 w/ 3DMark2005 (sec.) 283 358 342Wonder Twin Powers Activate! Form of benchmark!
DUAL-CORE PERFORMANCE PREVIEW50 MA XIMUMPC JUNE 2005
Best scores are bolded.