Undervolting 4450E + 780G, power usage results
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Undervolting 4450E + 780G, power usage results
I just built a diskless system using Gigabyte's 780 board and a 4450E:
AMD X2 4450E
Ninja B, fanless
Gigabyte GA-MA78GM-S2H 780G
integrated graphics, no graphics card
G.Skill 4GB (2 x 2GB) DDR2 800
Addonics SATA compact flash adapter w/16G 133x card
no hard drive
no optical drive
Antec NSK-1300
Nexus 120mm
pico PSU 120 w/110W brick
WinXP
Here are my power consumption test results, measured using Kill-A-Watt. I've listed the minimum stable undervolting I achieved, tested with 44 minutes or more of prime95 highest-power stress testing.
"off" power usage: 3W
Stock
------
2.3 GHz (stock 200x11.5)
1.25 V stock, reported by cpu-z: 1.264 V idle, 1.28 V load
idle: 32W
load: 77W
Undervolted, stock speed
--------------------------
2.3 GHz (stock 200x11.5)
1.025 V set in BIOS, reported by cpu-z: 1.04 V idle, 1.056 V load
idle: 28W
load: 57W
Undervolted, 2.0 GHz (this is how I'm going to use this machine)
----------------------
2.0 GHz (200x10)
0.95 V set in BIOS, reported by cpu-z: 0.96 V idle, 0.976 V load
idle: 27W
load: 49W
Stock Voltage & Clock, Cool-n-Quiet
-------------------------------------
1.0 GHz (200x5)
1.0 V stock, reported by cpu-z: 1.008 V idle, 1.28 V load
idle: 25W
load: 77W (cnq disengages at load, obviously)
Undervolted, 1.0 GHz
----------------------
1.0 GHz (200x5)
0.825 V set in BIOS, reported by cpu-z: 0.832 V idle, 0.832 V load
idle: 24W
load: 33W
Notes:
* I tried disabling IDE, 1394, serial and parallel, but this had no effect on idle or load power consumption.
* I did not try undervolting or underclocking anything other than the processor.
* I chose 2 GHz for testing because at this frequency the memory freq automatically assigned (CPU/5) is optimal (400 MHz). At all other frequencies, including stock maximum, the memory is clocked less than 400 MHz. Is it possible to get better control of this?
* Using this board it is not possible to set the multiplier higher than 11.5. Right? Is this an AMD thing or a Gigabyte thing? At frequencies > 2.0 GHz the memory is set at CPU/6. It'd be ideal to be able to set this processor at 2.4 GHz.
* For load power, I took a reading during the first prime95 test, so I wouldn't have to watch the meter for hours to catch it at its peak. I noticed, however, readings as much as 3W higher during other tests. I'm okay with this since prime95 is tougher than normal usage will ever be.
AMD X2 4450E
Ninja B, fanless
Gigabyte GA-MA78GM-S2H 780G
integrated graphics, no graphics card
G.Skill 4GB (2 x 2GB) DDR2 800
Addonics SATA compact flash adapter w/16G 133x card
no hard drive
no optical drive
Antec NSK-1300
Nexus 120mm
pico PSU 120 w/110W brick
WinXP
Here are my power consumption test results, measured using Kill-A-Watt. I've listed the minimum stable undervolting I achieved, tested with 44 minutes or more of prime95 highest-power stress testing.
"off" power usage: 3W
Stock
------
2.3 GHz (stock 200x11.5)
1.25 V stock, reported by cpu-z: 1.264 V idle, 1.28 V load
idle: 32W
load: 77W
Undervolted, stock speed
--------------------------
2.3 GHz (stock 200x11.5)
1.025 V set in BIOS, reported by cpu-z: 1.04 V idle, 1.056 V load
idle: 28W
load: 57W
Undervolted, 2.0 GHz (this is how I'm going to use this machine)
----------------------
2.0 GHz (200x10)
0.95 V set in BIOS, reported by cpu-z: 0.96 V idle, 0.976 V load
idle: 27W
load: 49W
Stock Voltage & Clock, Cool-n-Quiet
-------------------------------------
1.0 GHz (200x5)
1.0 V stock, reported by cpu-z: 1.008 V idle, 1.28 V load
idle: 25W
load: 77W (cnq disengages at load, obviously)
Undervolted, 1.0 GHz
----------------------
1.0 GHz (200x5)
0.825 V set in BIOS, reported by cpu-z: 0.832 V idle, 0.832 V load
idle: 24W
load: 33W
Notes:
* I tried disabling IDE, 1394, serial and parallel, but this had no effect on idle or load power consumption.
* I did not try undervolting or underclocking anything other than the processor.
* I chose 2 GHz for testing because at this frequency the memory freq automatically assigned (CPU/5) is optimal (400 MHz). At all other frequencies, including stock maximum, the memory is clocked less than 400 MHz. Is it possible to get better control of this?
* Using this board it is not possible to set the multiplier higher than 11.5. Right? Is this an AMD thing or a Gigabyte thing? At frequencies > 2.0 GHz the memory is set at CPU/6. It'd be ideal to be able to set this processor at 2.4 GHz.
* For load power, I took a reading during the first prime95 test, so I wouldn't have to watch the meter for hours to catch it at its peak. I noticed, however, readings as much as 3W higher during other tests. I'm okay with this since prime95 is tougher than normal usage will ever be.
Last edited by amyhughes on Mon Apr 28, 2008 9:28 pm, edited 2 times in total.
Re: Undervolting 4450E + 780G, power usage results
That's an AMD thing. For the 4450E, 11.5 is the highest you can go. Only the "Black Edition" CPUs have unlocked multipliers.amyhughes wrote:* Using this board it is not possible to set the multiplier higher than 11.5. Right? Is this an AMD thing or a Gigabyte thing? At frequencies > 2.0 GHz the memory is set at CPU/6. It'd be ideal to be able to set this processor at 2.4 GHz.
Is there an option to set your own memory divisor in the BIOS? (e.g. turning off memory timings "By SPD")
Are you going to use a program like RMClock to do performance-on-demand (like AMD's CnQ, except with lower voltages)? You could save some idle watts that way while being able to automatically ramp up performance when needed.
Re: Undervolting 4450E + 780G, power usage results
If there is I don't see it.davidh44 wrote:Is there an option to set your own memory divisor in the BIOS? (e.g. turning off memory timings "By SPD")
There's a memory controller freq setting that is only present with certain CPUs. For mine, it's not available, so I don't know what it does.
I was thinking since that CPU/5 was automatically reduced to CPU/6 over 2 GHz it'd automatically be increased at 1.6 GHz. I was expecting CPU/4 at 1.6 GHz and CPU/3 at 1.2 GHz. But it remains CPU/5 all the way down to 1 GHz.
It's going to run Linux, so unless there's a Linux version then probably not. In the configuration I'm going to run there's only a 3W idle savings to be had by reducing both frequency and voltage to minimum. Also, the machine will have a predictable load, so if I find 2 GHz is more than enough I can likely reduce freq and voltage myself.Are you going to use a program like RMClock to do performance-on-demand (like AMD's CnQ, except with lower voltages)? You could save some idle watts that way while being able to automatically ramp up performance when needed.
Well, actually UBUNTU "is" Linux, in the sense that UBUNTU is a distribution based on a GNU/LINUX OS called "DEBIAN"...amyhughes wrote:I haven't tested under Linux yet. I'm using Ubuntu.garacs1 wrote:Have you noticed any power consumption reduction switching from WinXP to Linux? Which distro are you using?
Anyway, I am using UBUNTU as well, so having a much older setup than yours, it is still even more interesting to see what figures you are getting from the kill-o-watt, to compare to mine. I have an "old" Pentium D820 on ASUS P5LD2 with 2gb Patriot high Performance DDR2 ram (2x 1GB kit) and an ASUS ATI X1600PRO VGA, and my kill-o-watt says I am sucking from the wall socket around 130W in idle (in windows, it's 150W).
What I meant is that the numbers I've posted are for Windows and I haven't re-run the tests for Linux. And in answer to your question, the distro I'm using is Ubuntu. Clearer, but oh so many wordsgaracs1 wrote:Well, actually UBUNTU "is" Linux, in the sense that UBUNTU is a distribution based on a GNU/LINUX OS called "DEBIAN"...
I don't think any of the power savings comes from Linux. In fact, I understand power management in Linux is not as good as Windows.Anyway, I am using UBUNTU as well, so having a much older setup than yours, it is still even more interesting to see what figures you are getting from the kill-o-watt, to compare to mine. I have an "old" Pentium D820 on ASUS P5LD2 with 2gb Patriot high Performance DDR2 ram (2x 1GB kit) and an ASUS ATI X1600PRO VGA, and my kill-o-watt says I am sucking from the wall socket around 130W in idle (in windows, it's 150W).
My current-gen main machine draws 105W idle. That's an E8400 system that is under-volted. I hope to get it down a bit with a smaller, more efficent PSU (doing that today), but I have no hope that system will ever approach my single-board, diskless AMD machine.
To get good dynamic undervolting in Linux, you need to roll your own kernel and modify the powernow-k8 module source
http://forums.gentoo.org/viewtopic-t-245197.html
http://forums.gentoo.org/viewtopic-t-245197.html
Very impressive results, thanks for sharing!
My brother and I also built a similar machine to your, with no fans or disk drives.
It uses the Gigabyte 690G motherboard with a Athlon X2 4800+ (stock heatsink without fan), 2x1 GB Kingston DDR667 and an IDE->CompactFlash adapter. It's in an AOpen HX08 case, with an older AOpen 350W PSU.
It runs fanless because of the BIOS undervolt to 1 GHz, 0.8v, and the heatsink is cool to touch. At idle it consumes 31W, and at load up to 38W. It also runs Debian 64-bit.
My brother and I also built a similar machine to your, with no fans or disk drives.
It uses the Gigabyte 690G motherboard with a Athlon X2 4800+ (stock heatsink without fan), 2x1 GB Kingston DDR667 and an IDE->CompactFlash adapter. It's in an AOpen HX08 case, with an older AOpen 350W PSU.
It runs fanless because of the BIOS undervolt to 1 GHz, 0.8v, and the heatsink is cool to touch. At idle it consumes 31W, and at load up to 38W. It also runs Debian 64-bit.
Don't have any experience with it of my own, but I found this link: http://www.thekip.nl/2008/04/06/quick-u ... linux-phc/jackylman wrote:Do you have the link?juamez wrote:I read something about the acpi-cpufreq module under Ubuntu that can be used for undervolting.