After one of my recent posts described how much I hated the wasted power outlet sockets under my desk due to the proliferation of power supply bricks, I ended up buying the Powersquid outlet multiplier as recommended by some folks, which did the job nicely.

Whilst I was there however, I couldn't resist also picking up a "Kill-a-watt" device.
(Both are available from Thinkgeek's electronics section).

Since this thing turned up, I've been obsessed with measuring power of things.

I use a dual head desktop setup at home, with a Dell 2001FP LCD as my first head, and an ancient 19" Acer 99SL CRT as my secondary head. Since measuring their power usage, I'm more keen than ever to invest in a new CRT.

Both running at 1600x1200, here's the stats..
Acer: 105W
Dell: 44W

Flipping to tty1, where I run 1024x768, something interesting happens..

Acer: 78W
Dell: 43W

Seems the LCD outputs more or less the same power regardless of resolution, whereas the higher resolution really pushes the CRT.

I also have another LCD, a 19" Acer that I attach to my firewall & IRC/web boxes via a dataswitcher. This one only does a maximum of 1280x1024, and like the other LCD, remains constant regardless of its resolution. It only sucks up a mere 33W though.

I thought the CRT was going to be the most power hungry thing I owned. Then I measured my Dell Precision 470 desktop, with its two thundering EM64T processors (with hyperthreading).

Sitting at an idle Gnome desktop, this thing sucks up 153W of power. I was almost scared to push it. But I did anyway.

Pulling Linus' latest git tree into my local tree sucked up 216W of power at its peak.

Just when I thought I'd seen it all, a -j4 kernel compile of that same tree pulled 258W.

I no longer leave this box on 24/7.


To give a comparison, my VIA C3 Nehemiah based firewall box runs happily day after day sucking down a mere 48 Watts. With the disk put to sleep a further 4 Watts are saved. Whilst this is vastly superior to the EM64T, I'm still puzzled why this is as high as it is. The CPU is supposed to pull a maximum of less than half that, and the only other components in the box are the chipset & memory.
I found it quite surprising it was this hungry.

48W isn't a lot though.
That's still less than the power needed to run my desk lamp (57W).
It's also the same amount as is needed to run my portable air con unit. (Which turned out to be a lifesaver this summer)

There's a handful of other things I have constantly turned on too as part of my home network.

16 port linksys switch - 3W
Motorola Surfboard cablemodem - 5W
Linksys WRV54G - 4W

And a few odds and ends that get turned on/off as needed.

4 port USB2 hub - 1W
usb hard disk - 11W
speakers (in standby) - 5W

Whilst I found these statistics quite interesting, I was very curious about how well our power management code actually works. So I did a bunch of tests on a HP nx1015v Athlon XP 1600+ laptop.

Idle in grub - 56W
Idle with apm (acpi=off) - 50W
Idle with no pm (acpi=off apm=off) - 50W
(Seems the APM implementation on this laptop is busted)
Idle at console (Using ACPI for pm) - 25W
Also using cpuspeed to lower CPU speed to 521MHz - 22W
Suspended to RAM (echo mem >/sys/power/state) - 3W
(Sadly, resume from RAM doesn't seem to work at all on this laptop).

Windows XP idle (Using ACPI afaik) - 20W
(Windows is probably doing more management of idle USB ports/PCI devices etc)
Windows XP suspended to RAM - 1W
(Again, Windows probably puts more stuff to sleep before we suspend)

Hitting the lid switch on this laptop drops the power usage by just 6W. (Surprising, given that aparently the screen is the most power hungry part of the laptop allegedly).

This gizmo has been quite fun to play with so far, and I've no doubt it'll continue to be useful for various power management projects.