Archive: SPCR's PSU Test Platform V.3

Power | Reference|Recommended
Viewing page 2 of 5 pages. Previous 1 2 3 4 5 Next


The PSU testing rig as described above has been in use since Q1 2004, nearly a year and a half. In that time, we've reviewed about two dozen power supplies. In general, we've been reasonably satisfied about the accuracy of our testing. However, there have been a couple of growing issues.


For some two decades, Intel has had the role of setting and driving technical standards for other companies to follow. It is part of the reason the PC industry grew as big and as quickly as it did. Without broad acceptance for literally hundreds of standards to define everything from the mechanical fit of components to guidelines for firmware, the industry would be very different than it is today. Intel's Form Factors web site and the various specifications authored by special teams at Intel are a visible sign of this standards-setting role. The Power Supply Design Guides for various form factors are just a few of many such standards.

The ATX12V Power Supply Design Guide grew out of the previous ATX PS Design Guide. It is not a mandatory standard, per se, but everyone knows that conformance with the current version of the Guide ensures unquestioned acceptance from other hardware makers and low risk of incompatibility.

In the time before ATX12V, much of the electrical power for the CPU was drawn by the motherboard from the 5V line. The first generation of Intel P4 systems changed this so that all the power for the CPU came from the 12V line, via the then-new AUX12V (2x12V) connector, which was defined and specified in the first iteration of the ATX12V PS Design Guide. For a while only P4 motherboards made use of the 2x12V connection, but in the last two years, AMD followed suit with its Athlon 64 (and related Sempron) processors, so that all 754, 939 and 940 motherboards draw power for the CPU from AUX12V. Today, the other major power hungry component in the PC, the VGA card, also relies almost entirely on 12V lines, whether through the motherboard or through direct connection to the PSU. PCI Express also draws mostly on the 12V line. The 5V line, once heavily used to power the CPU, now is called to supply power mostly to hard drives. The 3.3V line seems to be used mostly for RAM, PCI and AGP.

Each subsequent version of the ATX12V PSU Design Guide — 1.1, 1.2, 1.3, 2.0, 2.1, 2.2 — has increased the recommended 12V current. Version 2.0 introduced the concept of two 12V lines for higher power PSUs, with each 12V line not to deliver more than 18A. The +12V2 line is marked specifically to power only the AUX12V connector for the CPU.

What does all this have to do with SPCR's PSU testing system?

The DRAM DBS-2100 PSU load tester has the capability to provide a load of up to 23A on the 12V line. This was perfectly adequate with up to 400~450W PSUs conforming to v1.3 or earlier versions of Intel's ATX12V Power Supply Design Guide, but with the introduction of version 2.0 in mid-2004, 23A become inadequate for >400W power levels.

The ATX12V Guide provides typical power distribution recommendations for PSUs of different power output capability. In version 1.3, the most powerful model for which power distribution recommendations were provided was a 300W PSU:

+12V: 18A, +5V: 26A, +3.3V: 27A, -12V: 0.8A, +5VSB: 2A

Let's compare this to the current v2.2 recommendation for a 300W model and for a 450W model (the most powerful model profiled in the ATX12V Guide v2.2). Also noted are the maximum loads that the DBS-2100 PSU Load Tester can provide.

ATX12V Guide Power Distribution Recommendations
ATX12V Guide
v1.3 (300W)
v2.2 (300W)
v2.2 (450W)
DBS-2100 PSU Load Tester
Load Capability
For the electrically challenged, power (Watts) is obtained by multiplying the voltage (V) and current (Amps) for each line. The DBS-2100 is capable of loading a PSU to 276W on the 12V line (12Vx23A=276W) .

Looking first at the ATX12V Guide data, you can see that the 12V line is more highly stressed now than ever before, while at the same time, the reliance on the +5V and +3.3V lines has dropped dramatically. Compared to v1.3, the current recomendation is for less than half the power on the +5V line and about 2/3 on the +3.3V line. Converting the current (A) numbers to power (W) illustrates it more dramatically:

  • For a 300W model, Version 1.3 called for 216W on the 12V line, and a combined maximum of 195W on the +5V and +3.3V lines. (If all lines are drawn upon simultaneously, the total power would be limited to 300W.) A peak current capability of 19.5A on the 12V line (234W) for 17 seconds was also stipulated.
  • For a 300W model, Version 2.2 calls for a total of 252W on the 12V lines, and a combined maximum of just 120W on the +5V and +3.3V lines. (If all lines are drawn upon simultaneously, the total power would be limited to 300W.) A peak current capability of 25.5A on the 12V line (306W) for 17 seconds was also stipulated.
  • For a 450W model, Version 2.2 calls for 360W on the 12V lines, and a combined maximum of just 130W on the +5V and +3.3V lines. A peak current capability of 34A on the 12V lines (408W) for 17 seconds is also stipulated.

We recently published the article, Power Distribution in Six PCs, which confirms the high reliance on +12V in current PCs, up to 90% at full load in high power CPU systems, and the concomittant load reduction on the +5V and +3.3V lines. These were typically under 5A maximum.

The current capacities available on the various lines of the DBS-2100 PSU loader shows its age. It was built pre-ATX12V v1.3; you can see the enormous capacity available on the 5V and 3.3V lines, 195W and 102.5W, respectively. It is not capable of presenting more than a 23A load on the 12V line. This limitation could be worked around with v1.3 compliant PSU models, even those rated for much higher than 400W output power, because it was possible to max out the 12V line, then add more loading as needed on the +5V and +3.3V lines.

With the most recent v2.2 compliant PSU models, even testing at 400W is a bit of a challenge, and we've had to resort to maxing out the -12V and +5VSB lines to avoid overloading the +5V and +3.3V lines on the PSU. With some recent >500W PSU models, there was no choice but to overload all the non-12V lines on the PSU by a small amount in order to reach maximum load. This may have led to errors and unaccounted misbehavior.

TECH TIP: Even though there are two 12V lines in a >400W PSU that conforms to ATX12V v2.0 or higher, it is safe to wire these outputs in parallel. This is what's is done in the DBS-2100: All the 12V lines are electrically joined together. The reason this can be done is because the "independent" 12V lines are independent only in the sense that each line has a limiter which keeps the current to <20A. All the 12V lines actually originate from the same 120VAC:12VDC transformer and rectifier.


There has been a real and dramatic increase in PSU efficiency since SPCR first started measuring it in 2002. It was a pleasant surprise back then when a PSU managed to reach >70% efficiency at any load. The across-the-board >70% efficiency reached by the Seasonic Super Silencer 400 in mid-2003 was something of revelation, and no other PSU came even close to its maximum 78% efficiency until almost a year later.

Since the Enermax Noisetaker 475 broke the 80% mark in the spring of 2004, there have been at least 10 other PSU models that have reached 80% on the SPCR test bench. Almost all of these have been compliant with ATX12V v2.0 or higher, with much greater power on the 12V lines than on the others. This in itself was probably the source of a bit of the general rise in efficiency. The conversion from 120VAC to 12VDC is a more efficient process than the futher step down to 5V or 3.3V. One PSU engineer confided that a simple way to increase efficiency without making any substantial changes in a PSU is to rate the +12V line for slightly higher current and to derate the +5V and +3.3V lines for correspondingly lower current. The end result will be an efficiency rating that could be 2~3 percentage points higher.

The increases in efficiency also coincides with another aspect of the ATX12V Guide's evolution. Until version 1.3, the Guide included only the most cursory note about efficiency; it was required to be at least 68%. In v1.3, the required minimum was raised to 70%. In version 2.0, the minimum efficiency was recommended to be 75% at full load, 80% at typical load and 68% at light load. This was far more specific and demanding about efficiency than Intel had been before.

In version 2.2, the required minimum remains at 70%, but the recommended guidance is even higher: 77% at full load, 80% at typical load and 75% at light load. Typical and light loads are now defined in the same way that 80 Plus defines it: 50% and 20% of full rated power. In fact, Intel has adopted the entire PSU load testing protocol used by the 80 Plus program. A PDF copy of this document, Internal Power Supply Test Protocol Rev. 4.0, can be downloaded from the web site

With Intel setting higher recommendations, 80 Plus urging >80% efficiency and SPCR extolling the potential cool and quiet virtues of a high efficiency PSU, perhaps it is no surprise that manufacturers have responded by making more efficient PSUs. Still, there were questions about the several PSUs we've tested recently that reached >85% efficiency. If 80% is such an easily reached efficiency target, why aren't there more 80 Plus approved models?

Our AC/DC conversion efficiency results have almost always tended to be a bit higher than the specifications published by the manufacturers of the tested PSUs. It seemed only a small discrepancy at first, but as the rated maximum power and 12V current capacity of tested units increased, so did the discrepancy.

The Seasonic SS-400HT Active PFC F3, which we tested in August, was one that had passed the 80 Plus requirements for >80% efficiency. 80 Plus had tested a sample of this model, and their result showed a high of 85%, compared to 90% in our results. This was a real cause for concern, as several months of previous dialogue with the 80 Plus team had convinced me of their careful, professional approach to PSU testing. Why were we getting such different results?

To allay or confirm our growing suspicion that something was amiss, we conducted a complete examination of all the factors and procedures that go into obtaining efficiency in our test system. This included:

A. AC Power meters. All the various Kill-a-Watt and Seasonic Power Angel AC power meters were checked against each other and against readings off a couple of multimeters. As far as we could tell, the AC power meters appear to provide very good accuracy for power, typically within 1% of each other.

B. The Question of PSU Voltage Drop. We had noted but never accounted for the drops in voltage that occurs at >50% load with most power supplies. The worst case is usually at maximum power, where the 12V line might drop to 11.80V, for example, which is still well within the 5% tolerance specified by the ATX12V Guide. However, at high load, a slight drop on the 12V line can mean more than a couple of watts. Looking at the data collected for the Seasonic SS-400HT, however, we found the 12V to be dead on at full load. There were small drops on the +5V and +3.3V lines; these were examined and found to cause a total drop of ~5W at 400W output. This means that instead of 400W, the unit was actually delivering ~396W. It led to a 1% drop in efficiency, down to 89%. That was not enough to explain the discrepancy between the 80 Plus test results (83.2% at 393W load) and ours.

C. Accuracy of the DBS-2100 Load Tester. This was an issue we dreaded facing. Was the PSU Loader giving us loads that were lower than indicated by the markings on the front panels switches? If a switch marked 8A on the 12V line was acutally only forcing the PSU to draw 7A, for example, this would be 12W less power output than expected. A few miscalibrated switches like this would certainly cause our efficiency results to be inflated erroneously.

There was only one way to test this: Measure the DC current drawn on each output connector, systematically going through each of the relevant 27 switches across the DBS-2100 control panel. We used a reasonably high quality Fluke 36 clamp ammeter for this job, and also measured the voltage at the output terminals with a digital multimeter for each and every test. The current meter was checked for accuracy against another more expensive meter; accuracy seemed good, as the readings were generally within 0.1A of the other meter. Three different power supplies were used to check on whether interaction between the PSU and the load tester affected the output current.

A Fluke 36 Clamp Meter was used to measure the current for every load setting, including innumerable combinations. It is shown above measuring a current of 4.3A flowing across the +12V line (+12V1) of a Molex output connector.

This exhaustive testing took over two days. At the end of the testing, we had some sobering answers to several questions about the DBS-2100, which did turn out to be the main source of our efficiency error:

1. Does the PSU deliver the current as marked on the PSU loader switches? Not always. In many cases, the delivered current is lower than marked. It becomes worse at higher loads. It is not a linear across all the power levels, the error varies with total power load and voltage lines.

2. How bad is the error? At lower loads, there is no error, or it's just 1~2% low. As the load is increased, it can get as bad as 6% low, at the highest loads with certain combinations of switches on certain voltage lines. Going back to our example of the Seasonic SS-400HT, it turns out that the output power was not 400W as we thought, but just 378W if we assume all the voltage lines were precisely at 12V, 5V and 3.3V, which they were not. Adjusting for the ~4W reduction due to the slight voltage drops on the +5V and +3.3V lines, the actual power delivered was 374W. Recalculating efficiency (374 ÷ 443 x 100%) we get 84.4%, which is very close to the 83.2% at 393W output reported by 80 Plus on their sample of the same PSU model.

3. Is the error consistent with different power supplies? Generally, yes, although there are variances. At our standard test load distribution settings for 400W, for example, we obtained 367W, 374W and 375W, respectively, from the three PSUs.

Previous 1 2 3 4 5 Next

Reference|Recommended - Article Index
Help support this site, buy from one of our affiliate retailers!