Also, since the power supply found on the PC have five different outputs (+12 V, +5 V, + 5 VSB, +3.3 V and –12 V) you would need to connect five multimeters to the power supply at the same time, and publications using this methodology usually use only one, measuring the outputs in different moments in time, making the results worthless, since they were taken in different moments, with different conditions (load, temperature, etc). Even if you connected five multimeters, you would need to read them at the same time. We don’t know any human being able to read five instruments at the same time and write down the values at the same time. Even if you are really fast, you will take some seconds to make this measurement. As we already mentioned, things inside the PC power supply happens in microseconds, so seconds make a huge difference.
One way to use the above methodology correctly is by using a device to grab the value of all five outputs at the same time, like a digital data collector. The problem is that we would measure the voltages, which, once again, mean nothing. One way to make a correct power supply test using this approach is by measuring the current (and not the voltage) of the five outputs at the same time using a data collector, if you add a correct load to the power supply. In fact, this methodology is the one suggested by one engineer at Intel and can work out if you have the right equipment. We will describe this idea on the next page.
Another problem regarding the use of regular multimeters is precision. We cannot guarantee the precision of low-cost multimeters. If you add five multimeters, we cannot guarantee if the multimeters are calibrated among them, showing the exact same results when measuring the same thing.
Load Testing
The second major problem with almost all PSU reviews around is the use of an inadequate load.
Some websites will use a regular PC on their PSU review. The problem is that high-end power supplies nowadays can deliver at least 600 W and a regular PC is not able to pull all this power. Even if you use a very high-end PC with two CPUs, several hard disk drives and even four video cards, you won’t be able to say how much power you are pulling at a given moment and you also won’t be able to say up to how much power you were able to pull from the power supply, since you are not using any measurement device.
A methodology some publications use is setting up a passive load, by connecting the power supply to a series of wire resistors. With this approach you can say how much current and thus power the power supply is delivering by applying ohm’s law (I = V / R or P = V^2 / R), as the voltage and resistance are known. Even though this is an interesting idea, there are some problems with this approach as well. First, you will need to build a resistor net for each output to be used at the same time (we’ve seen some publications adding a resistor net to only one of the power supply’s output or to all outputs but at different moments, not at the same time). Second, you can see some smoke coming from the resistors, as they will heat a lot and they can burn (some publications will add a fan over the resistors). Third, this is a passive load test and computers are dynamic systems, meaning that they can pull a lot of power at one moment and then reduce the power consumption some seconds later and then increase the power consumption again after some more seconds. With a passive load like this you also won’t be able to measure automatically several important features like efficiency and protections.
One word about efficiency. Efficiency is the relation between how much DC power the power supply is being able to deliver and how much AC power it is pulling from the power grid to deliver it. With the above methodology this can be calculated if you add an AC power meter to the power supply AC input. As you will have the amount of DC power the power supply is delivering and the amount of AC power the power supply is pulling from the power grid, efficiency will be given by the formula DC power / AC power.
The best way to review power supplies is using an active load tester, like Chroma 8000. This machine will accurately measure the maximum current and power a power supply is capable of delivering and is also able to measure several other parameters and features, like all power supply protections. The only problem with this tool is that it costs almost USD 50,000.
There are some load testers on the market that internally have just a series of wire resistors, so the commentaries about resistor load are also valid for such machines.
We mentioned the methodology suggested by one engineer from Intel, Andrew Watts. His approach is to add a current sensor to each power supply output and connecting them to a digital oscilloscope and to a data collector, then connecting the data collector to a PC running monitoring software. This is a very interesting approach, however the problem with this methodology is that you still need a PC to generate the system load. If you want to take a better look on this methodology – which can be cheaper than buying a Chroma machine –, click here to download a PDF presentation about it. We are also showing below some pictures of the board created by Andrew to connect the current sensors between the power supply and the PC (the BNC connectors are used to connect the digital oscilloscope) and also some pictures of the system assembled and running.
Notice that the oscilloscope and the computer connected to the data collector is measuring the current consumed by the PC (and thus power, thru the formula P = V x I), not the power supply voltages.
click to enlarge
Figure 1: Board with current sensors.
click to enlarge
Figure 2: Board connected between the power supply and the PC used as load.
click to enlarge
Figure 3: Data collector, digital multimeter and digital oscilloscope.
click to enlarge
Figure 4: Software running (it is measuring the power consumption).
As you can see on Figure 4, the software can tell the power consumption for each output, the total power consumption and also the power supply efficiency.
Conclusions
We have a motto here on Hardware Secrets: if we can’t do something correctly, we prefer not doing it at all. Between publishing a lousy review and not publishing it at all, we prefer not publishing it. That is why we do not “review” power supplies: what we do here is to disassemble power supplies and make an in-depth analysis of the components and project used, as you can see visiting our Power Section. We are saving money to buy a Chroma machine, and we hope to buy one soon. As soon as we buy one, we will start publishing “real” reviews, together with the in-depth analysis we already do.
We hope you have learned to detect a good power supply review, and you will be amazed in noting that almost all reviewing websites are using wrong methodologies that really tell us nothing about the power supply real power capabilities. Even worse, some websites are recommending bad power supplies based on flawed methodologies.
If you are a reviewer, don’t get us wrong. Our purpose here is to educate both users and reviewers, so you now have a better knowledge on power supplies and have learned what you should not do. One suggestion? Instead of calling your articles as “reviews” call them something else, like “First Impressions”, “First Look”, etc, if you are not using a real load tester.
We should point out that there are some websites doing a terrific job on power supply reviews. SPCR (SilentPCReview), for example, goes one step further, adding a variable AC power supply, which allows them to simulate different power grid configurations – not only to check if the power supply can really operate under the range stated by the manufacturer (e.g. 90 V – 240 V) but it also simulate spikes and noise on the power grid. They also use a load tester (even though is a simpler model compared to Chroma 8000).
Other websites worth mentioning are Xbitlabs, which created their own active load for testing PSUs, Planet3Dnow, a German website that owns a Chroma 6330 machine and also used a Chroma 8800 in some reviews, and JonnyGuru, that also uses a load tester (a Sunmoon SM-8800). There are more websites around using a load tester, these are the ones I am most familiar with.
[
本帖最后由 flight 于 2007-5-10 22:08 编辑 ]
