frame rates when running on faster vid-
eocards, as well. It’s likely that when Half
Life 2 is released, we’ll dump either Jedi
Academy or Halo (described next).
For our Jedi Academy test, we created
a custom demo that uses the Tatooine
map with tons of bots running about the
environment. We also used “Force Seeing”
and “Force Push” a number of times in
order to push physics calculations. The
demo is run at 1280x1024 with 32-bit
color and 32-bit texture quality; trilinear
filtering, high-geometric detail, and tex-
ture detail are all set to “very high.” In
the videocard’s Display Properties control
panel, antialiasing is turned off, but we set
visual quality settings to the maximum.
The audio engine is enabled, and the final
score is reported in average frames per sec-
ond (higher scores are better).Halo
Gearbox’s Xbox-to-PC Halo port is a seri-
ous system ball-breaker. For the last two
years people have been complaining that
no games really punish PC hardware; now
people are complaining that Halo is too
system-intensive. The DirectX 9 game
makes use of pixel shaders that conform
to the brutal Pixel Shader 2.0 spec, and
seems to scale well as both CPU and vid-
eocard power increase.
For our test, we use the game’s built-in
timedemo mode with the resolution set to
1280x1024. We set in-game visual qual-
ity settings to high for the demo run, and
leave audio enabled. In the videocard’s
Display Properties control panel, antialias-
ing is turned off, but we set visual quality
settings to the maximum. The final score
is reported in average frames per second
(higher scores are better).Meet the New Test Beds
When we update our benchmarks, we
usually update our test bed machines as
well. These five identical hand-built PCs
are used as host systems for testing com-
ponents reviewed in the magazine. The
test beds are also used to establish the
“zero-point” numbers referenced in our
benchmark charts (see sidebar).
Each of our current test beds is armed
with a 2GHz Pentium 4, an Intel D850MV
mobo, 512MB of PC800 RDRAM, a
GeForce4 4600 Ti videocard, a WD
1000BB hard drive, and a Sound Blaster
Audigy. This config is a quaint antiquity
by today’s standards, but it was hot 18
months ago, and we believe that even a
power user shouldn’t have to do a com-
plete system overhaul more than once
every 18 months. Plus, because our zero-
point systems must be as stable as possi-
ble for component testing, we sometimes
err on the side of caution. For example,
we went with a conservative Intel moth-
erboard because we trusted it to present
the fewest number of hardware conflicts.
The question of speed versus stabil-
ity became a thorny issue during this
month’s zero-point refresh, because we
were forced to choose between Intel’s and
AMD’s latest CPUs—each a worthy com-
petitor in its own right.
On one side, we had Intel’s Pentium
4 Extreme Edition (P4EE), which offers a
very stable platform. Boards with Intel’s
875P chipset have been available since
June, and any known core-logic kinks
have long been ironed out. Plus, Intel’s
drivers and chipsets have historically been
bulletproof.
On the other side, we had AMD’s
Athlon 64 FX. Back in the day, we
wouldn’t have touched any third-party
chipset designed for an AMD CPU, at least
not when laying the foundation for our
test beds. Conflicts between soundcards
and motherboards have been a problem
for AMD’s partners, and chipset driver
updates were coming out so fast and
furious, you had to check for new ones
every week. Today, however, we
don’t have any pressing con-
cerns about the reliability
of AMD chipsets, and
we’re also very
impressed
with theAthlon 64 FX’s on-die memory controller
and the promise of a 64-bit OS.Let’s Config!
In the end, we decided to build our new
test beds around AMD’s latest platform,
specifically, the Athlon 64 FX-51. This
processor is a bit slower than the P4EE in
regular apps, and tests on par with the
P4EE in games. Plus, the FX-51 will allow
us to test 64-bit apps when the 64-bit
version of WinXP arrives, and the proc is
available in quantity right now. If we had
chosen the P4EE for our five test beds, we
would have likely run into serious avail-
ability problems.
This latest test bed refresh marks
the end of a seven-year reign for Intel
in our Lab. Don’t let us down, AMD!
We’re counting on you for a stable infra-
structure, which brings us to our next
configuration question: Which chipset?
Unfortunately, we haven’t had enough
time to extensively test VIA’s K8T800 and
nVidia’s nForce3 Pro 150, the two core-
logic chipsets designed for AMD’s new
CPU. So, for our motherboard decision,
we simply went with our gut instincts,
which led us to the nForce3-lovin’ Asus
SK8N. Scuttlebutt has it that VIA’s chip-
set offers slightly more performance, but
nVidia’s chipset has been on the market
longer, so we’re hoping it’s more mature.
Plus, Asus has a rep for making some of
the most solid mobos around.
For main system memory, we went
with 1GB of registered DDR400 from
Corsair Micro, a company that’s known
for both its high-performance RAM and
a predilection to not sell clock speeds
it can’t deliver. On the video front, we
selected the ATI Radeon 9800 Pro XT ,
which offers industry-leading pixel-shad-
ing performance, a stable driver base, and
excellent visual quality in a majority of
available games.
When it came to our hard drive deci-
sion, we ran into a quandary. As of
press time, the nForce3 Pro 150 chipset
didn’t offer native Serial ATA support.
Our motherboard, the Asus SK8N, does
support Serial ATA, but only through an
onboard Promise controller. This presents
two problems. First, onboard Serial ATA
controller chips route data through the
133MB/s PCI bus, and we didn’t like the
prospect that Serial ATA data—which
has a top speed limit of 150MB/s—might
some day bottleneck in the slower PCI
bus. Second, when we run component
testing, we do numerous system reformats
and OS installs, and we weren’t looking
forward to installing new Promise drivers
every time we even breathed on one of
our test beds.
In the end, we decided to go with the
250GB Western Digital WD2500 —aIn The Lab
updates were coming out so fast and
furious, you had to check for new ones
every week. Today, however, we
don’t have any pressing con-
cerns about the reliability
of AMD chipsets, and
we’re also very
impressed
with theHere they are, folks—five gleaming new
Athlon 64 FX-51 CPUs patiently waiting
for deployment in our test beds.continued from page 93continued on page 96
94 MAXIMUMPC DECEMBER 2003