How We Test
Our new benchmarks stress real-world applications
over synthetic performance
JUNE 2006 MA XIMUMPC
W
e’ve always preferred real-world tests to measure
PC performance. By using common games and
applications, we feel that our tests give readers a more
realistic picture of a machine’s abilities. It’s no secret that
vendors tune their drivers and tweak their hardware to
meet the demands of various benchmarks. It’s a great
thing if they’re tuning for real-world apps, but it has little
value to the consumer if they tune for a synthetic bench-
mark—and in the past, tuning for synthetic tests has even
hurt performance in games and apps.
Our latest system benchmarks adhere to our “real-
world” philosophy, and even kick it up a notch. Trying to
fi nd real-world tasks that stress the capabilities of a dual-
core SLI rig isn’t easy. Sure, we could run some ridiculous
3D render or esoteric test that calculates pi to 100,000
decimal points, but these don’t have much practical
value. We wanted to address chores that truly are issues
for real users, such as editing high-defi nition home movies.
(HD cameras might be too pricey for a lot of folks now, but they’ll
soon be under $900, so expect to see more high-def cams at
Disneyworld.)
We’ve also added a few more tasks to our Photoshop test,
and we start with a RAW fi le (we used to begin with a high-resolu-
tion JPG fi le) to simulate what a digital photographer might do.
Additionally, we compress a movie for playback on a PSP, using
Nero Recode ; and measure OpenGL and DirectX gaming perfor-
mance at fairly high resolutions.
We think our new benchmarks are a good measure of the perfor-
mance a typical Maximum PC should deliver.
SYSmark2004 SE
Some people might call SYSmark2004 SE a “synthetic” test, but
they’d be wrong. If anything, SYSmark2004 SE is as real-world as
you can get. The benchmark suite uses 19 mainstream applications
to simulate real-world offi ce and content-creation tasks.
Unlike tests that isolate and measure the time it takes to com-
plete a single task in a single application, SYSmark2004 SE mea-
sures the time it takes from the issuance of a command until the
test app actually does something and the system responds. In other
words, SYSmark measures how long you spend waiting on your
computer. Both the 2004 and SE versions contain very few multi-
threaded apps (beyond Discreet’s
3D modeling application 3dsmax ),
but the program runs several apps
at one time, so dual-core CPUs per-
form better than single cores.
We’ve used the previous ver-
sion, SYSmark2004 , for many
years. The SE version adds sup-
port for 64-bit OSes. (Unfortunately,
fundamental changes to the app
make it impossible to compare
scores from SYSmark2004 SE and
SYSmark2004.) SYSmark2004 SE is
calibrated based on the performance
of a 2GHz Pentium 4 Northwood
system using an Intel 845 chip-
set, 512MB of DDR266, an 80GB
7,200rpm IBM hard drive, and an ATI
Radeon 9700 Pro. This confi guration
achieves a score of 100. Our zero-
point system’s score of 275 therefore
indicates that it’s 2.75 times faster
than a 2GHz P4 box. Thanks to
its well-rounded nature, SYSmark
stresses the CPU, hard drive, RAM,A score of 275 in BAPCo’s SYSmark2004 SE means the perfor-
mance is almost three times that of a 2GHz Pentium 4 box.Our new zero-point systems and benchmarks are the perfect complement to our spacious new Lab. Continued on page 6 4 Ë