AnandTech's power consumption testing methodology

[John]

One thing that people need to keep in mind is that not everyone is using the same testing methodology. While I appreciate Anand's articles, I have a hard time digesting the full load benchmarks in some articles since there are inconsistencies. They do appear to be measured at the wall with a kill-a-watt meter. In the future I would like to see the AnandTech editors list the particular psu they are using in the test system specs chart. This will allow people to get a better idea of the true load on the psu.

We tested power consumption for these parts in the same way we usually do, by measuring the total power draw of the system with each of the cards installed in two different states. The first state is with the system idle (no other programs running) and the second is while the GPU is under stress testing. We use a few of the benchmarks from 3DMark06 to stress the GPUs to find out their power consumption under load. Because we are measuring the wattage of the entire system and not simply the cards, we can only get a general idea of the type of power levels of these cards.

Take a look at these two AnandTech articles.

[*]Intel Core 2 Duo E4300: Affordable and Highly Overclockable

[*]GeForce 8800 Roundup: The Best of the Best

Considering the test beds are similar, and the X6800 draws 70-75W, why is there such a gap in the measured wall loads?

Running 3DMark will not yield a true full system load. You also need to be running a high resolution, HDD read/writes, and TAT or Orthos at the same time to determine a true full load IMO. Please take into account that the article(s) you refer to may or may not be accurate enough to support your own claims when you are recommending psu's and discussing power consumption.

 

[PurdueRy]

Originally posted by: John
One thing that people need to keep in mind is that not everyone is using the same testing methodology. While I appreciate Anand's articles, I have a hard time digesting the full load benchmarks in some articles since there are inconsistencies. They do appear to be measured at the wall with a kill-a-watt meter. In the future I would like to see the AnandTech editors list the particular psu they are using in the test system specs chart. This will allow people to get a better idea of the true load on the psu.

We tested power consumption for these parts in the same way we usually do, by measuring the total power draw of the system with each of the cards installed in two different states. The first state is with the system idle (no other programs running) and the second is while the GPU is under stress testing. We use a few of the benchmarks from 3DMark06 to stress the GPUs to find out their power consumption under load. Because we are measuring the wattage of the entire system and not simply the cards, we can only get a general idea of the type of power levels of these cards.

Take a look at these two AnandTech articles.

[*]Intel Core 2 Duo E4300: Affordable and Highly Overclockable

[*]GeForce 8800 Roundup: The Best of the Best

Considering the test beds are similar, and the X6800 draws 70-75W, why is there such a gap in the measured wall loads?

Running 3DMark will not yield a true full system load. You also need to be running a high resolution, HDD read/writes, and TAT or Orthos at the same time to determine a true full load IMO. Please take into account that the article(s) you refer to may or may not be accurate enough to support your own claims when you are recommending psu's and discussing power consumption.

Most likely because they are stressing the CPU in the first with Prime95 or similar program and, as stated, 3dmark06 in the second.

In the first they want to compare CPU power consumption and in the second mostly GPU power consumption. It's not a perfect method but it helps take out the factor not under test.

 

[dclive]

Agreed. And even then, even with the least efficient 8800GTXs, you're still right around 300 watts used. But yes, I agree it's odd - one chart shows a 55W jump, and another shows a 100W jump, when systems are put under a load.

Net net -- you're still around 300W required. Call it 350W if you get the peltier-cooled 8800GTX.

 

[Zepper]

If those powers are read at the wall with a KaW, then what is actually being drawn from the PSU will be reduced by the efficiency of the PSU at that output point. The efficiency would reduce the PSU power output by at least 15% to as much as 35% relative to power from the wall. IOW, in the 8800 comparo, that 301 watt figure from the wall means the PSU is putting out 255 Watts max with a top efficiency PSU like an 80 Plus certified Seasonic model, to as little as 227W with a more typical 75% efficient PSU.

.bh.

 

[SerpentRoyal]

Power at the wall is the instantaneous product of line voltage and line current. That's the most accurate method of measuring power used by the PSU. The actual power used by the rig is often 15 to 25% lower, depending on the efficiency of the PSU at that particular load. Yes, efficiency can vary based on PSU load and temperature!

Orthos/Prime does not adequately tax the CPU. Best program for this type of test is TAT or S&M HEAT mode.

 

[Howard]

Originally posted by: SerpentRoyal
Power at the wall is the instantaneous product of line voltage and line current.

VA is, anyway.