Introducing The Beast
When we set out to revisit power supply coverage, we realized we needed to do more than just plunk a PSU into a test system for some quality time at idle and under load. Testing in a real-world system is essential, of course, but it doesn't necessarily push a power supply as hard as we'd like. We want to see how a power supply performs when stretched to its limits. Those limits, of course, are the maximum output wattages printed so conveniently on the side of every power supply. Reaching them requires the ability to generate arbitrary loads on a power supply's various voltage lines. Since load generators for PC power supplies aren't exactly off-the-shelf retail products, we set about building one from scratch.
My brief flirtation with electrical engineering is long behind me, but TR regular justbrewit has helped us with electrical projects in the past, and he was up to the challenge. JBI even wrote up his experiences building our power supply tester, which we've shared below.
Many moons ago, Geoff contacted me with an idea. I believe the origin of the idea was actually a thread on the TR forums, where I had commented on a device he had put together to help measure the power consumption of hard drives. In a nutshell, TR was interested in a device to help them test (and stress-test) PSUs.
The trick is to be able to apply controlled loads to each rail in a repeatable way, and push PSUs to the limits of their rated wattage to see how they behave under stress. With PSU wattages rapidly climbing past the 1 kilowatt mark, this becomes a non-trivial problem; the little off-the-shelf PSU testers you can buy do little more than check whether the voltages are within reasonable limits, under very light (just a few watts) loading.
We kicked around various concepts, including a bank of adjustable high-power rheostats. Eventually I proposed a design to Geoff which was based on banks of binary-weighted load resistors, with individual switches for each load. After a number of tweaks to the initial proposal to accommodate future PSUs (we added provisions for up to four +12V rails), and reduce cost and complexity (support for testing +5VSB and -12V rails was dropped), we had a pretty good idea of what we wanted The Beast (as it came to be known) to look like.
The final specs: 0 to 46A on the 3.3V rail, 0 to 62A on the +5V rail, and 0 to 88A on the +12V rails (four 22A rails which can be ganged together), all in calibrated 2A steps. That's over 1500W of combined load surely enough to handle all but the most extreme PSUs! A pair of 120mm Sunon fans (powered from their own +12V wall wart power brick) keep The Beast from losing its cool.
After numerous delays, parts were finally ordered and construction got underway this past January. My basement bathroom was temporarily turned into a makeshift machine (and soldering) shop, since it was too cold to work out in the garage. Geoff shipped me a pair of OCZ PowerStream PSUs to help with testing and calibration -- since The Beast is designed to handle over 1000W on the +12V rails alone, I had to gang two conventional PSUs together to test the tester, and ensure that it had adequate cooling!
Further delays ensued when I discovered that the main ATX connector on The Beast's wiring harness was prone to bent pins, meaning it was not rugged enough to stand up to repeated plugging and unplugging. Some re-engineering of the wiring harness to ensure that the pins remain straight in the housing, plus making the main ATX connector modular (so that it is easily replaceable) took care of that issue.
During testing and calibration, I also managed to destroy a ThermalTake TR2 PSU, when an incorrect switch setting caused me to accidentally exceed the PSU's combined maximum wattage rating. A popping sound, a flash of light (visible through the PSU's exhaust fan opening), a puff of smoke, and that was all she wrote. Beast indeed!
In order to see how each PSU reacts at its limits, we'll be using The Beast to test them at 50, 75, and 100% of their rated capacities. Our tests will load the 3.3, 5, and 12V rails simultaneously, so we have to keep in mind each PSU's combined and total power output limits. Those limits dictate our power draw targets at 50, 75, and 100% capacity, and each PSU's individual rail biases govern how the load is distributed across the 3.3, 5, and 12V rails.
In May, the finished Beast was carefully packed, and finally shipped off to Geoff. A few days ago he informed me that it performed quite well during the testing he did for the PSU roundup... and he even managed to avoid blowing any of the PSUs up!
Since The Beast is limited to applying loads in 2A increments, we won't be able to nail the percentage-based load targets exactly. Instead, we've channeled The Price is Right and used amperage loads that come as close to our targets as possible without going over. The chart below shows the amperage loads applied to each rail during testing.
|Total loads (Amps)|
|Antec EarthWatts 500W||6||6||14||10||10||22||14||14||30|
|Antec Neo HE 550W||8||8||16||14||12||24||18||16||32|
|Antec TruePower Trio 650W||8||8||20||14||14||30||18||18||40|
|Cooler Master Real Power Pro 550W||8||6||16||12||10||24||18||14||34|
|Cooler Master Real Power Pro 650W||10||10||20||14||14||30||20||20||40|
|Corsair HX 620W||6||8||20||10||12||30||14||16||40|
|Enermax Infiniti 720W||6||8||24||10||12||36||14||16||48|
|OCZ GameXStream 700W||8||6||22||12||10||34||16||14||46|
|PC Power & Cooling Silencer 750W||6||8||24||10||14||36||14||18||50|
|Seasonic S12II 500W||6||6||14||10||10||22||14||14||30|
|ThermalTake Toughpower 700W||8||8||22||12||12||34||18||16||46|
When testing with The Beast, each power supply was hooked up using its primary and auxiliary 12V connectors, two PCIe power connectors, and six 4-pin peripheral connectors. We used a Pico ADC-212 digital oscilloscope to probe 3.3 and 5V wires on the primary power connector. 12V lines were probed in the primary power connector and also with one of the PCIe power connectors. In the graphs on the following pages, 12V power from the primary connector will be marked 12V1, while power from the PCIe connector will be 12V2.
At each load level, we logged DC and AC voltage for 100 seconds after an initial warm-up period where the PSU was under load for five minutes. DC voltages were averaged over that period and tended not to vary much, if at all. Each line's AC content, otherwise known as ripple, varied quite a bit, as the following example shows.
The graph looks a little scary, but ripple is normal for a power supply. You just don't want too much of it, and in the example above, the spikes don't reach 30 millivolts. Because layering the ripple results in multiple lines would make our graphs entirely too difficult to read, and since you probably don't want to scroll through a dozen individual graphs for each of the 11 PSUs we're testing, we've elected to present ripple content as an average of the absolute value of AC voltages over our 100-second test period. Think of this measure as the average amplitude of the AC ripple results, using zero as a baseline. Ideally there should be no AC content in a DC line, so an average of absolute values should give us a good representation of just how far each PSU strays from that ideal.
In addition to probing each PSU with an oscilloscope, we also used a Watts Up? PRO meter to log power draw at the wall socket. These wattages were averaged across our 100-second test interval, although the results rarely deviated by more than 0.1W during that time. These socket draw results were used to determine each PSU's efficiency under 50, 75, and 100% loads.