Power Consumption - What is your footprint? Part 1

Add Your Comments

HTPC Power Consumption Part 1

In the following article (Part 1), I will go into some detail over the true power consumption of certain devices. This does not only apply to Home Theatre PC’s, but to every PC when you have a thought for the environment, or your wallet. Both are just causes for concern.

First and foremost, I will be conducting most of the readings using my Kill-A-Watt meter. I have seen quite a few questions raised over the accuracy of Kill-A-Watt. I am no electrician, or even understand other than the basics of electrical circuits of any kind so I am merely repeating the findings of others. As far as I have read, when compared to other scientific digital power analyzers, the Kill-A-Watt meter is well within accuracy to use as a device even as high as about 1000w. As heavy load increases, the accuracy of the meter will play a larger role, but for the majority of readers, this will have no effect on you or the scope of this article. For those who are using over 1000w, I believe if you are concerned over power consumption down to smaller tweaks and peripherals, you may wish to rethink your choice of components altogether.

I plan to go through an array of both BIOS and hardware configurations that will measure wattage and compare before and after trials to see “what” exactly makes the biggest difference in power consumption. I will also throw in temperature differences, as I would consider that part of power consumption if this affects users in smaller rooms where A/C may need to be slightly increased. All temperature readings were obtained at the same time and day to maintain accuracy due to ambient temperature fluctuations in my house. Accuracy should be within a margin of error of +/-2° c.

So let us start with hardware as this should logically make the largest difference in power consumption.

The first test involves processor choice. For this test I will be using the following setup with nothing left out of the list seeing as every piece of hardware counts.

Software to be used for considering “load” will be Prime95 Large FFT’s for 1 hour. Idle consumption will be determined after a fresh restart and left alone for 1 hour with no processes using anything more than 1-2% CPU usage.

Setup:

CPU 1a: AMD AM2+ X4 620 @ 2.6GHz stock voltage (45nm 95w Quad Core)
— Under a fanless Ninja Mini with 2x 90mm and 1x 120mm fans as shown above

Motherboard: DFI Lanparty Jr 790GX-M2RS

GPU: Ati Radeon HD4550 512MB – stock

RAM: 2x1GB DDR2 PQI Turbo DDR2-667

Drives: 5 Hard Drives – 2x 320GB Seagate 7200.10’s in RAID1, 1x 1.5TB Seagate 7200.11, 1x 500GB WD Green, 1x 400GB WD Cavalier – all 7200rpm. 1x Pioneer DVD Burner

PSU: Corsair VX450

Case: Silverstone LC20B-M ATX HTPC Case. Small VFD display enabled.

My HTPC example

My HTPC example

** The Beginning… **

To begin, let’s start with processor usage. In Part 2 of this article (coming soon), I will discuss an X2 3600+ CPU and a BE-2300, of which will show how die improvements have affected consumption. Stay tuned for this. Now back to the X4 620.

X4 620 @ 1.3v BIOS
–1.04v idle: 124w @ 35c
–1.3v load: 208w @ 59c

** Undervolting **

I managed to undervolt my X4 620 from a 1.3v stock down to 1.07v keeping her prime95 stable. Each voltage measurement is obtained via CPUz. At idle, with Cool & Quiet enabled, she was using a mere 0.77v which is quite remarkable. Here are the following readings after a number of tests:

@ 1.3v BIOS (Set to auto)
–1.04v idle: 124w @ 35c
–1.3v load: 208w @ 59c

@ 1.025v BIOS (Set to -0.275v. Undervolted stable setup)
–0.77v idle: 118w @ 33c
–1.07v load: 160w @ 44c

As you can see above, there is a significant reduction in both temperatures and power consumption by adjusting something as simple as the CPU voltage when under load. Even more interesting, but of course not as satisfactory, was only a minor drop in consumption at idle even though voltage was reduced by 0.3v between the two tests. There must be some reason as to why there is only a small change…Possibly Cool & Quiet functions better on automatic?

To help give a better answer to this, I decided to run a 3rd test in which I set the VID to only a small -0.025v decrease, or -25mv to see if when set to “auto” causes results to skew.

@ 1.285v BIOS (Set to -0.025v)
–1.01v idle: 123w @ 35c
–1.29v load: 201w @ 56c

As you can see, we can confirm Cool & Quiet is NOT the cause. I believe it is safe to assume for idle solutions, there is not too much difference, but the increase in heat remember can cause fans to run higher for longer (not applicable in this scenario) which would emphasize the difference further.

Undervolting Conclusion:
It is a great idea to undervolt to take a few watts off the idle consumption (5% decrease) and a little off the idle temperatures (2° C decrease), but the real magic shined under load as tests revealed a 30% decrease in power consumption and a 15° C drop in temperatures. If you have no plans for overclocking and a long term outlook for your computer, this is an absolute must (in my opinion). All hardware will last longer due to lower ambient temperatures in the case and you can start to make a better impact on both the environment and your wallet.

** ACC / L3 cache **

In this short section, I will see if the addition of L3 cache affects consumption of the processor. My X4 620 was lucky enough to have the L3 cache included when ACC (Advanced Clock Calibration) was enabled. The benefits of L3 cache are clear; most articles I have read show a 5-15% increase in applications and games, but are the power requirements and/or temperatures affected?

BIOS Baseline Read / ACC OFF: 119w idle @ 32c – 3dmark2006: 3702 – 200w load @ 55c
BIOS Changed Read / ACC ON: 121w idle @ N/A – 3dmark2006: 3715 – 206w load @ N/A

Temperatures with ACC ON cannot be recorded due to the temperature monitoring bug becoming active when it is turned on. Load temperatures were established using Prime95, not 3dmark.

ACC/ L3 Cache Conclusion:

It seems that there is barely any increase in idle consumption, and only a small increase on load with this enabled. There are a number of reviews out that show benefits to the L3 cache in synthetic benchmarks are true and it can help performance slightly, so I won’t go through and re-bench what has already been done. Enable it if you do any number crunching or gaming! It is definitely worth it.

** Overclocking RAM**

Next up I wish to measure the difference that different RAM frequencies and voltage make in regards to power consumption. This would a great way for you to play devils advocate to see if a small increase in consumption is worth X% benefit in performance.

The tests are very simple and the results are below for my 2x1GB PQI Turbo DDR2-667 set at idle with my CPU undervolted 0.3v — the same conditions as in the undervolting section for the processor.

BIOS @ DDR2-400 (1:1, 200MHz effective) 1.925v = 118w
BIOS @ DDR2-533 (3:4, 266MHz effective) 1.925v = 118w
BIOS @ DDR2-667 (3:5, 333MHz effective) 1.925v = 118w
BIOS @ DDR2-667 (3:5, 333MHz effective) 2.220v = 121w
BIOS @ DDR2-667 (3:5, 333MHz effective) 1.800v = 116w

Overclocking RAM Conclusion:
Negligible if any increase in consumption when voltage is not touched, so if you can find evidence for faster operation in your computing duties, then go for it. Even when voltage was bumped to just over 2.2v there was only a 3w increase, so again, if your performance increases bump your speed as much as you want. When it comes to undervolting, it can take off a few watts, so it is worth it for consumption and longevity of the hardware.

This concludes Part 1 for now. In part 2 of this article (coming soon), I will cover 2-3 more processors (x2 3600+, BE-2300 and possibly a Q6600 stock and at 3.6GHz as a heavily overclocked example), overclocking graphics card clocks (2d vs 3d, idle and load), the consumption of modern day LCD’s while running and in power savings mode and finally hard drives.

If there are any other items that you would love to see included in part 2, feel free to shoot me an email at the bottom and I will see if I have the hardware capacity to run the tests!

ps2cho

Leave a Reply

Your email address will not be published. Required fields are marked *

Discussion
  1. You might want to adjust to actual load power or DC_W since silentpcreview has reviewed the Corsair VX450, and does list its efficiency from AC_W = 35 to 563W.
    Right, but as I understand it, that makes no difference when it comes to the amount you $$ in your bills, right? It is just a "theoretical". Correct me if I am wrong.
    DC_W gives you the real or actual trend vs power (DC_W). AC_W distorts the trend since efficiency is usually much lower in the low power regions...of interest in low power/low heat/low noise/HTPC/etc.
    For example, a system may show 5-10 W_AC difference at the wall while actually a 10-20 W_DC difference. A larger spread that is masked by the inefficiencies of a switching power supply at low loads.
    The few data points in your editorial looked like they would be in a 80+% efficiency range, but I assume you have more data, especially for the other processors.
    gimpster123
    nice start, waiting for part 2. I'm curious to see how some of the "power consumption calculator tools" for choosing a PSU measure up to actual tests.

    They are heavily overstated. In reality you can cut whatever they "require" by about a 1/3.
    computers don't really need much power compaired to other appliences in your house
    these are mine
    my computer uses 460W
    My Aircon uses 1286W
    my TumbleDryer uses 1987W
    joshhua5
    computers don't really need much power compaired to other appliences in your house
    these are mine
    my computer uses 460W
    My Aircon uses 1286W
    my TumbleDryer uses 1987W

    But how long do you use your dryer for? It's not something that is on all day long.
    Even an A/C usually turns on and off in cycles dictated by your thermostat.
    So you have to weigh the power usage over the period of time you are using the device. Many leave their computers on 24/7.
    460w is not an unreasonable amount of power consumption for a computer but I'm guessing in your case you are assuming it uses 460w because you have a 460w powersupply? That is simply not how it works. For one, the wattage rating of the power supply is for output, not input.
    My power usage is pretty crazy (rig in sig, using kill-a-watt):
    Idle at desktop: 375watts
    Prime95 (4 CPU cores loaded): 530watts
    Furmark (4 GPUs, 1 CPU core loaded): 877watts
    Prime95 (3 CPU cores loaded) & Furmark (4 GPUs, 1 CPU core loaded): 1035watts
    Prime95 (3 CPU cores loaded) & Furmark (4 GPUs, 1 CPU core loaded) & Defrag on all non-SSD Hard Drives: 1173watts
    I am usually between 675-800w while playing most games.
    As GNR mentioned, it is inaccurate to refer to energy as power (rate) as you were doing.
    In my region, electricity is billed per KW-Hr--1000 W for 1 hr, 100 W for 10 hr, 1 W for 1000 hr, etc.
    My UPS says my computer pulls about 225w while I'm browsing.. Never saw over 335 while playing badlands. That is with the monitor.. Doesn't look like it's using SLI.. and apparently doesn't need it. lol I'll have to do more looking later.
    BrandonB
    My UPS says my computer pulls about 225w while I'm browsing.

    The UPS pulls more power, or much more depending on the newness of the design and its maximum load. You'll probably want to be in the 50 to 100% load range.
    BrandonB
    Yes, I realize that.. however I'm using the UPS's LCD to monitor the output.. my kill-a-watt isn't currently hooked up. :)

    You mean the UPS can monitor its own output, the output of the power supply, or both?