Archive: SPCR's PSU Test Platform V.3

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

THE SOLUTIONS

Our findings call for a complete revision of our testing procedure for efficiency and loading. It does not change the validity of the other tested PSU parameters such as noise, fan controller behavior, voltage regulation, or thermal performance in any significant way. The details learned about the DBS-2100 suggests we should be monitoring the current individually for every setting and voltage line, and factoring in any sags in PSU output voltage into the efficiency calculation. Or perhaps it's time for a better more accurate PSU testing device.

Pursuing the latter option first, I contacted some PC power supply engineers at Antec, Intel, Seasonic and Fortron. I found that the most widely used PSU testing rigs are those made by Chroma ATE. Inc. The reason for the popularity of Chroma PSU testing systems is that for their level of accuracy and testing automation capability, they are cost-effective. Unfortunately, cost effective for a PSU manufacturer is not the same as cost effective for SPCR. Even used, a suitable Chroma testing rig would set us back at least US$5,000. The unit would most likely require updating and changes if Intel changed its ATX12V Guide substantially.

A complication of this option is that there is no easy way to feed the heat of the load back into the PSU thermal simulation box or to rig up something similar in function. Furthermore, the Chroma and other advanced PSU testing equipment all have substantial fan cooling, which cannot not be disengaged. This means noise testing would suffer; noise testing of the PSUs requires low ambient noise levels, which can be obtained with our current test rig. The sheer cost and the various drawbacks associated with the Chroma option made it impossible for us to consider seriously at this time.

So... We went back to our existing test system and considered the changes that would make it viable for more accurate testing results in the context of today's 12V-heavy power distribution. The objectives we identifed:

1. The load distribution on the voltage lines at the various power output levels should reflect the proportional ratings of each line of the tested PSU. This is the methodology used by the 80 Plus program, and also by Intel's PSU design / test engineers. A PDF copy of the protocol, Internal Power Supply Test Protocol Rev. 4.0, can be downloaded from the web site www.efficientpowersupplies.org. It's a matter of a bit more math for each PSU load test, rather than using fixed load settings for each line based on power output level as we have done up to now. This change will make our results more directly comparable to those obtained by 80 Plus and other organizations.

2. Increase the 12V load capacity. An additional 20A load capacity for the 12V line would be very useful.

3. Measure the actual current and voltage for each line individually for each power level rather than trust the current value marked for each switch on the DBS-2100 loader, and use these measured numbers to determine actual DC power output. Use this data to calculate efficiency. This means a longer, more tedious process of data collection when running PSU tests, but it will ensure that the inaccuracies of the DBS-2100 loader never enters the data stream.

GOAL #1 is achieved simply by following the referenced documentation closely. However, rather than test at 20%, 50% and 100% loads, we have decided to retain our previous power output levels of 40W, 65W, 90W, 150W, 200W, 250W and 300W. This is the critical range where most systems operate 95% of the time. For PSUs rated for higher power, we'll go up beyond 300W by 100W or 150W increments to full power.

GOAL #2, increasing the 12V load capacity, required more attention and more work. There was much discussion about various tools and options. In the end, my choice was dictated by availability, cost and ease of implementation.

SWITCHABLE RESISTOR BANKS

I fabricated a bank of 12 20W resistors into a series / parallel network to provide five individually switchable loads of approximately 1.7A, 1.7A. 3.2A. 6.4A and 6.4A at 12V, and any combination thereof, up to a total of 19.4A. The network was wired to a 4-pin 2x12V and an 8-pin 4x12V connector so that either the AUX12V or EPS12V output could be used.

As the resistors would have to dissipate up to 230W, I organized the network into two banks. Each bank of six resistors are clamped between a pair of heavy aluminum plates that act as heatsinks. There is enough space between the resistors to allow airflow between them. The photos and captions below will explain better than words alone.


One of the resistor banks, placed on bottom aluminum plate. The wiring was recycled from old, dead PSUs.

Thicker, heavier top plate atop the resistors.

The resistors are clamped between the aluminum plates, with screws raising the structure for airflow below. To ensure good thermal conduction, a small thermal interface pad was placed between the resistors and the plates. This material acts much like TIM goop for CPUs and heatsinks, filling gaps and evening the contact.

The finished resistor banks are at the bottom of the PSU simulation box, in the flow of air from the fans in the DBS-2100 load tester. There is an inch of space behind each bank so the airflow can pass through between the resistors. All wiring connections are soldered.

Note 5-switch "front panel" with 2x12V and 4x12V connectors. Unfortunately, an error was made with the latter ° it's a "male" 4x12V plug that was laboriously soldered in place, but what's required is a "female", which I have not been able to locate since discovering this error. Ah well... at least it does no harm.


How the DBS-2100 and PSU thermal simulation box go together.

The end result is that the PSU can now be loaded up to ~23A on +12V1 using the DBS-2100, and up to ~19A on +12V2 using this new additional resistor bank, for a total of about 42A, or over 500W. All this on just the 12V lines. The load banks are entirely independent, so the current and power to each can be reported separately.



Previous 1 2 3 4 5 Next

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