Numbers don’t lie. Or do they? The
advertised wattage rating is the one figure that every computer user
looks at when choosing a new power supply. The problem is, it’s a
figure you can’t really trust. I found a PSU advertised as an 1,100W
blockbuster, but it couldn’t run a cutting-edge PC without randomly
rebooting. At the same time, some so-called 610W and 620W units powered
the same system without any problems (see the following article for
write-ups on those). Unfortunately, few manufacturers market
their power supplies the same way. Some sell PSUs by the continuous
wattage ratings the units can pull 24/7; others, by their peak power
draws that shouldn’t be sustained for more than a minute. Some vendors
arrive at their ratings at testing temperatures of 35 degrees Celsius
(95 degrees Fahrenheit) or even 25 C (77 F)—unrealistically low figures for a high-end computer’s case temperature—while
others rate their PSUs at 40 C (104 F) or a more grueling 50 C (122 F).
It does make a difference: A PSU that can handle 1,000W at 25 C may
only be good for 850W at 50 C. Still others, like the 1.1kW
unit I mentioned above, are simply overprotective, shutting down
whenever anything electrical goes slightly astray. It’s admirable that
protection of your computer components is foremost in some
manufacturers’ minds, but what about the disk errors from unexpected
shutdowns? Fortunately, most PSUs strike a balance between stability
and protection. All of these considerations are becoming
crucial now that power requirements have become crazy-bad in top rigs.
Dual-core processors gave way to four cores. Two-drive RAID 0s are
nervous about three-and four-drive RAIDs 5 and 6 on PCI Express. And
Quad SLI GeForce 7950 GX2s may soon be bumped by Quad CrossFire R600s,
plus physics cards. When you’re shopping for an expensive
power supply to run that ultimate PC you’ve built, you don’t want to be
short-changed. Here’s my take on a large group of hopefuls from 700
continuous watts on up. How I Tested Confusion
reigns in PSU marketing, so I set out to clear the air with some harsh,
real-world tests. I mounted and ran each unit in a closed case—not on an open-air bench—under stressful operating conditions. I used a test system with the following configurations: A: 2x FX-74, 2x 8800 GTX SLI (8XAA/16XAF) B: 2x FX-74, 1x 8800 GTX (8XAA/16XAF) C: 2x FX-74, 1x 8800 GTS 640MB (2XAA/8XAF) D: 2x FX-74, 1x 7900 GTX (2XAA/8XAF) All
configurations featured the following: 3GHz processor speed, 4x 1GB
Corsair CM2X1024-8500C5D Dominator DDR2, two Raptor 10,000rpm hard
drives, one WD 7,200rpm drive, two Sony DVD drives, one floppy drive,
Asus’ L1N64-SLI WS (nForce 570 SLI) mainboard, ForceWare 97.92 drivers,
and a Dell 3007WFP display at 2,560 x 1,600. Systems C and D didn’t have enough video RAM to run 8XAA/16XAF at 2,560 x 1,600, but 2XAA/8XAF worked. I used Windows XP SP2—fully patched—as
I didn’t want any instability attributed to incompatibilities between
my software and Windows Vista. PC Wizard 2006 reported the system
temperatures. Not every power supply could handle a top-end
system, so I varied the graphics card configuration to get a rough idea
of each unit’s capabilities. Torture-testing a PSU via this method is
basically like using a go/no-go gauge. You want a power supply that
will easily handle your current system build and, ideally, your next
one, too. But it will either work stably under a heavy, continually
changing load, or it won’t. If a PSU didn’t have enough 6-pin PCI-E
connectors, I used adapters from one of the video cards’ retail
bundles. I continuously looped 3DMark06 1.1.0 as individual
instances of Prime95 ran on all four CPU cores using the In-Place Large
FFTs Torture Test. The idea was to maximize the test PC’s power draw by
utilizing as much of the CPU and GPU(s) as possible, with the RAM, hard
drives, and motherboard core logic getting a workout along the way.
3DMark also provided variation in load to better test transient
response. I noted the beefiest configuration each PSU could reliably
run between two to four hours (I’d like to have tested each unit for
several days) in the “Tested Stable To” row of the accompanying chart. In
configuration A, this truly hungry system sucked a maximum of 865W
under load depending on the power supply, according to my Extech 380803
power analyzer. You know, your basic PSU’s nightmare. I
hadn’t planned for my computer case to add to the PSUs’ (and my own)
torture, but it did. Thermaltake’s Armor Extreme doesn’t let you pull a
power supply out the back like any sensible case; you have to move it
through the front drive bays. Even more egregiously, Thermaltake
decided to turn the PSU on its side and put a hard drive cage next to
it. This not only places the weightiest components in the upper rear of
the case (although it does allow the hard drives’ fan to evacuate heat)
but also makes fan-side power harnesses several inches shorter. Most
damningly, it causes the hard drive cage to block any bottom-mounted
PSU fan that would normally face the CPU(s). On the bright side, this
setup allowed me to test affected PSUs under horrible ventilation
conditions. One thing I dropped from my testing was direct
measurement of the voltage level ranges. Like the Nvidia 680i Evga
motherboard I’ve used earlier, my Asus nForce 570 SLI board didn’t
report its +12V and +5V voltage levels. Also, my inexpensive digital
multimeter exhibited voltage variation even at baseline, so I’d have to
call my hands-on measurements inconclusive. In the end, I reasoned that
any components running too far out of spec would make themselves known
through instabilities. I certainly didn’t want anyone to base a buying
decision on such questionable data.
PC Power & Cooling Silencer 750 EPS12V Quad
Silencer 750 EPS12V Quad $199 PC Power & Cooling
www.pcpower.com
CPUs: 4 PC
Power & Cooling is very pointed about why it believes its power
supplies are superior. Founder Doug Dodson built industrial power
supplies for harsh environments for nigh on two decades before turning
his expertise to the consumer PC market, and it shows in the company’s
products. Big capacitors on the input side can better survive
momentary power losses and brownouts (holdup time). Big capacitors on
the output side offer stability under sudden load changes, such as when
F.E.A.R. gets graphically intense (transient response). A rear fan
doesn’t take away space for big heatsinks like a bottom fan would.
Inductors and capacitors carry high temperature ratings, such as 105 C.
All PCP&C units are tested in a 40 C oven at a 110% of maximum load
at 90V instead of 110 to 120V, which means the PSU has to draw more
current (and endure more heat) to supply the same wattage. Thus,
it’s not surprising that this PSU supports “all CrossFire setups,
including new products that are shipping this spring.” I confirmed that
all R600 variants are included in this pronouncement. At $199, this is
a best buy indeed.
PC Power & Cooling Turbo-Cool 1KW SR
Turbo-Cool 1KW SR $499 PC Power & Cooling www.pcpower.com CPUs: 4.5 It’s
probably the best you can get, but you’ll pay for it. At half a grand,
this PSU retails for more than most of us have ever spent on a graphics
card. Heck, I’ve seen entire PCs for less. This single-rail
über PSU is long, both in case dimension and generous cable length. If
you don’t have 11.5 inches between the back of your case and your
bay-mounted drives, don’t bother. The company throws in some 6- to
8-pin PCI-E adapters, so you won’t have to worry should your wallet
have enough cash left over for an R600. There are cheaper
alternatives with more connector flexibility. Many of them may offer
service that’s just as reliable and long-lived. But if you want ample
headroom for overclocking and future upgrades (not to mention
world-class stability), consider this unit an investment.
Watt, Me Worry?
PC Power & Cooling’s 1KW SR is the top dog here, but you can’t rule out the Enermax Galaxy DXX 1000W and Cooler Master Real Power Pro 1KW for $130 to $180 less. For power on a budget, get the PCP&C Silencer 750 and never look back.
by Marty Sems
|