F
or years, striping together two or
more hard drives in a RAID 0 array
has been the power-user’s configura-
tion of choice. Using two drives instead
of one, with both drives sharing the
workload, can result in an astounding
performance increase over a single drive
thanks to the drives’ ability to execute
tasks in parallel. Benchmark numbers
have backed this up; sequential read
speeds typically jump from 50-odd MB
per second for late-model SATA drives to
more than 100MB when configured in
RAID 0.
Lately, however, two well-known
hardware sites ( Anandtech.com and
Storagereview.com ) have come down
hard on RAID 0’s desktop performance.
This summer, both sites posted stories
claiming that, for single-user workloads
and gaming, these gains in data transfer
rates are either nonexistent or so mini-
mal as to not be worth the extra cost
and risk levied by RAID 0. Obviously,
this contention flies in the face of
conventional PC-enthusiast orthodoxy
(notice that all six gaming rigs in this
month’s Desktop Deathmatch feature
use RAID 0), creating a firestorm of
discussion within the PC hardware com-
munity. In the interest of peer review,
we decided to investigate further with
our own extensive battery of Lab tests.
Our findings: RAID 0 provides per-
formance benefits over a single drive,
but primarily in the realm of sequential
transfers such as video editing or other
apps that require more bandwidth.
RAID also delivers improved scores in
the Application Index of the hard drive
benchmark H2benchw —a test based on
real-world usage patterns for a desktop
user. But what about the loading of
huge maps in 3D games? That would
surely be faster with RAID, right?
To find out, we coupled a 3.2GHz
Pentium 4 with an Intel 865/ICH5
chipset, to which we connected in turn
a single boot drive, a bootable RAID 0
array, a nonbootable single drive, andnonbootable RAID array using
the same drives. The tests con-
sisted of launching a new single-
player campaign in three different
games — Far Cry , Doom 3, and
Battlefield 1942.
Much to our surprise, we found
minimal performance benefits
from our RAID array. And, in
some cases, we even saw a slight
decrease in performance! This ver-
dict remained consistent regard-
less of the configuration we used.
The startling results prompted
us to extend our tests to a bevy
of different drives with varying
performance capabilities, includ-
ing an external FireWire drive, a
4,200rpm laptop hard drive, a PCI
RAM disk, a 7,200rpm USB drive,
and even a 15K SCSI drive. We
also tried both a host-based SATA
RAID controller card and a parallel
ATA RAID card with an onboard
CPU. RAID or no RAID, in almost
all instances, game-level loading
times hardly wavered.
How is this possible? We sus-
pect (and our test with the RAM
disk seemingly confirms this) that
today’s games are far more CPU depen-
dent in terms of mission and level load-
ing. This means that more time is spent
decompressing levels, textures, and
sounds than reading from the disk. To
determine just how important the CPU
is in loading levels, we switched from
a 3.2GHz Pentium 4 to a 2.0GHz chip,
and saw the level load time increase by
14 seconds!
So why does everyone swear RAID
feels faster? All we can say is that
seat-of-the-pants computing
sometimes flies in the face of
facts. We were taken aback by the
results, but we’ll freely admit that
the RAID world is round even
though most of us thought it
was flat.
As it stands, RAID 0 is anexpensive way to obtain a small perfor-
mance advantage, but this has always
been the case. As stated earlier, we
were genuinely surprised by our game
loading tests, and have tempered our
enthusiasm for RAID 0 in light of these
results. Rest assured, we’ll continue
to investigate ways desktop users can
maximize the performance of their stor-
age subsystems and will bring you more
information in future issues.In the Lab A behind-the-scenes look at Maximum PC testing
The Case Against Raid 0
An Internet brouhaha surrounding RAID 0 performance spills
into the Maximum PC Lab—and produces shocking resultsMA XIMUMPC OCTOBER 2004
Running several hard drives in RAID 0 boosts
sequential read speeds, but Maximum PC Lab
tests indicate that for games, using a single
drive is usually faster.expensive way to obtain a small perfor-
mance advantage, but this has always
been the case. As stated earlier, we
were genuinely surprised by our game
loading tests, and have tempered our
enthusiasm for RAID 0 in light of theseRunning several hard drives in RAID 0 boosts
sequential read speeds, but Maximum PC Lab
tests indicate that for games, using a single
drive is usually faster.sounds than reading from the disk. To
determine just how important the CPUenthusiasm for RAID 0 in light of these
results. Rest assured, we’ll continue
to investigate ways desktop users can
maximize the performance of their stor-
age subsystems and will bring you more
information in future issues.enthusiasm for RAID 0 in light of these
results. Rest assured, we’ll continueA behind-the-scenes look at Maximum PC testing
DARE TO COMPARE: Game Level Load Times
RAID ARRAY SINGLE DRIVE
Doom 3 (secs) 38 35
Far Cry (secs) 21 21
Battlefield 1942 (secs) 22 22
* Maxtor DiamondMax 10 hard drives were used to perform these tests.