Testing the Supermicro-PWS 0065

Detailed test – Super Nade & JonnyGURU

In these days of power hungry components, the benign ATX power supply is beginning to play an increasingly important role.

The power-supply can make or break an enthusiast’s dream (and his/her wallet). What exactly would constitute an enthusiast’s PSU? I would like to set the bar as high as possible, so a typical unit should be able to handle an x1900xtx CF or a 7950 SLI setup on a Prescott based system. I chose the Prescott as it is the epitome of power inefficiency. If the PSU can power a Prescott based dual GPU system, it can easily handle the current generation of CPU’s.

To meet these demands, among other things, a single, high ampere 12V rail is recommended. JonnyGURU has an excellent introductory series of articles which go into this matter in detail – I would highly recommend everybody to read them, as he has provided the data and testing for this part of the review.

A few contenders at the time of acquiring this unit were Zippy (PSL-6600P), Silverstone (Zeus 560W) and OCZ (Powerstream 510 SLI). Zippy units would break my bank, the Silverstone had already been reviewed extensively and the OCZ would not suffice, according to several reports on the web. I decided to get this unit on the recommendation of Galvanized at Badcaps.net.

Today, we will take a look at the Supermicro-Ablecom PWS 0056, a server unit manufactured by Lite-On Inc. This unit satisfies, on paper, the requirements of an enthusiast PSU. The reason I opted for this unit should be pretty apparent, 46 A on the +12V rail. Looks very good on paper!

The unit itself looks pretty military – no fancy lights or shiny finishing – it looks rather severe. Of course, one can not expect any embellishments in a server unit. In fact, this unit does not have an ON-OFF switch!

The cables have no sleeving or any accessories – typical server unit:

Now we come to the fun part! The primary side caps are made by United-Chemicon, a reputed manufacturer of low ESR devices. The goop in there is to damp physical vibrations. Physical vibrations (manifesting as a hum) occurs when harmonics of the line frequency are produced.

Mathematically, any waveform can be decomposed into sine and cosine waves (the basis of Fourier analysis) and since the audible range extends from 20 Hz – 20 kHz, harmonics, if produced, can be heard. You can see that the inductor is not wound properly. Usually, one avoids crossing windings. It looks to be a solid build. The green caps are Taicon.

From the ATX Design Guide:

“Ripple and noise are defined as periodic or random signals over a frequency band of 10 Hz to 20 MHz. Measurements shall be made with an oscilloscope with 20 MHz bandwidth. Outputs should be bypassed at the connector with a 0.1 µF ceramic disk capacitor and a 10 µF electrolytic capacitor to simulate system loading.”

Let us see how this unit holds up to these numbers. First off, I’d like to say that this unit does not confirm completely to the ATX standard, as it has more than 18A on the 12V line.

Please visit JonnyGURU to understand his testing procedure!

• First off, it is easy to see that this unit has poor power efficiency – I expected something on the lines of 85%.
• In both the Cold and Hot tests (please refer to JonnyGURU’s site for a detailed description), the 12V line has held very well. The maximum deviation was just about 1.5% (under the hot test), which is well below the 5% limit. This is remarkable considering the fact that supplies rarely maintain their rated efficiency or regulation at high temperatures
• JG said the fan made a loud noise when the temperature increased. Note that this unit has a redundant backup fan if the primary exhaust failed. I don’t see any advantage in this – they could have used a 120 mm or a dual fan configuration instead.
• In the hot/full load tests, JG reported a buzzing sound from the inside of the unit. I would guess that this is due to a small valued secondary cap which is not filtering out higher harmonics. Also, we discussed the possibility of the toroidal inductors acting up. If you look carefully, the windings are horrible, with plenty of criss-crossing. Usually in an SMPS device, there is always going to be a magnetic hysteresis effect, which can result in an audible hum as energy is dissipated. This can be exacerbated by a poor quality inductor.
• The cross-load test result is really bad. I would chalk this up as average (11.45V is the failure mark). What does this mean? Since all rails interact with each other for some measure, unbalancing load on one rail would affect the others. In our case, lightly loaded 5.0V and 3.3V rails and a heavily loaded 12V rail yields a near failing result. An easy way to circumvent this issue would be to have about 4 HDD’s and an Optical drive, plus lots of fans.

• JG mentioned that the 12V line drops by about 0.1V and immediately recovers “2ms sharp drop”. This could again be chalked up to undersized caps. I was also thinking in terms of Ground Bounce due to small ground loops or if it is step like, it may have something to do with the fly-back of the voltage regulator. I mean, during a switching cycle, the Volt x time product is not being maintained momentarily. If it is a rapid exponential, maybe the secondary caps are too small in value?

The X axis is 2ms / div and Y axis (vertical) is 0.05 V/div, here one division corresponds to one square. Each square is further subdivided into 5 sub-units, so that makes each sub-unit 10 mV in the Y axis.

Now, let us look at the final test where all rails are almost fully loaded.

Ripple on the 12V line stays at about 60 mV while putting out 40 A! Holy cow, this is even lower than the ATX spec of 120 mV! This works out to a regulation of 0.5%. This is quite remarkable indeed! I would be more than pleased with such performance! All this is from a $136 unit!

Solely based on this data, I would say that this unit would be a superb buy. However, I would ALWAYS see how the unit performs in a real-world system before recommending it without any reservations. But so far, this unit is the champ in my stable.

This particular unit offers superb value for money. Based solely on the automated test results, I would tentatively say that this unit is all set to occupy the niche in the power-hungry SLI/CF segments. I will provide a definitive answer once I finish testing it on a real-world system.

However, there are several things which work against this or any other server class unit. Most significant among them deal with aesthetics and noise. This unit comes with no power switch, no cable sleeving, a mess of cables, and no SATA or SLI connectors. However, if you need the juice, who gives a damn about outward appearance anyway?

Super Nade overclockers.com/forums
JonnyGURU, jonnyguru.com