Testing Penryn power consumption, P vs T, and TDP.

Hey, I have started testing out some processors for power consumption, but I need some help. Here are a few questions I have:

1. I have been using BatteryBar and RMclock to provide with "instantaneous" power usage figures for the laptop. Does anyone know how this is determined? Is there a built-in ammeter, or is this calculated some other way? How reliable is that?

2. I have been letting the processors idle with the screen at minimum brightness, no underclocking nor undervolting, the hard drive idle but powered on, and the wireless card turned off. I save a screen shot of the lowest power consumption figure I see at idle, then I run orthos and save a screenshot of the highest power consumption figure I see at full load. Does anyone have a better idea for comparing power consumption between processors? Doing it by temperature is a bad idea since the fan turns on and off to keep the temperature within a range and I can't control it on my laptop.

3. I want to clock processors at all the same clock speed and the same voltage for another part of the test, but I can't seem to be able to control it with RMclock. The options for engaging Intel Dynamic Acceleration and enabling Dynamic FSB Frequency Switching are grayed out and I am unable to select them. I'm running vista 32-bit on my laptop. This would help tremendously.

Any help or suggestions on this project would be appreciated.

And last but not least, a teaser pic:

 

why not just look up the power numbers on intel.com? Their measurement equipment and techniques are going to be far better/accurate than anything you will be able to rig up.

 

The best possible measurements you could take would be the power usage from the wall. Get a Kill A WATT.

Good luck on this, but you are going to need to make sure you keep your tests consistent between all the processors. Use the same computer, same OS, turn off all prefetching/superfetching/defragging/etc that you can, lock the power profile to keep the CPU running at full speed and/or minimum speed, etc.

There are definitely better ways to measure the power consumption, but those techniques require hardware modifications best left to Intel.

 

Thanks Greg, that Kill A Watt sounds interesting. I've run through 4 processors so far and have kept everything the same except I forgot the fetching stuff. I made this thread because the results so far just seem so weird.

If this information was available, I wouldn't go through the trouble.

 

I'd be interested in this - Intel says the TDP (approximately the maximum power consumed) is 25W for P, and 35W for T, but that's about it. That appears to be assuming the maximum supported voltage is being run as well - so, for my T7500, the 35W should be at 1.3V with IDA running at 2.4 GHz. But in terms of how much power a processor will use at idle - it's kind of a void. And I've long wondered whether the T series and P series actually are that different at idle. Sure, the P series should win at load (before undervolting in considered - then, it may depend, even on the individual chip), but if you want maximum battery life, you want to know idle power consumption as well. And if T uses almost identical to P at idle, P isn't much of an advantage there.

I have no idea why IDA and DFFS and grayed out for you. They were selectable and enabled on my RMCPUClock from the beginning (running XP 32-bit). As far as I know, they were on XP 64-bit and Vista 32-bit as well, although I never payed attention to those boxes before, and no longer have either of those operating systems intalled. Perhaps it's due to your specific system?

 

You're going to need to keep many variables constant, which will be quite difficult. I would suggest either somehow making sure you have disabled all background processes such as System Restore, Windows Defrag, Superfetch, any antivirus/firewalls, etc etc, or doing a clean install (but even then, it would require tweaks).

To check your tests are still valid, you're going to need to test certain processors twice or more. Say you start off with a T9400 as your first processor. Make sure you test it again sort of as a control once around the middle of your other tests and again at the end, to make sure no other variables are messing with your results (ie: a clean install becoming more "stable" with less HDD thrashing later on).

I agree with Greg that a hardware solution would give you better, more "independent" results.

This will be a difficult experiment in which to control all the variables, but that said, I am interested in seeing your results.

 

the best possible power numbers would be from the fet power components inside of the laptop. The best possible temp readings would be from two temp probes of known calibration placed directly above and below the cpu die.

Anything else is going to have way too many variables to be credible.

The power and temp numbers for the cpus are indeed available on Intel.com. Look.

 

I'm also interested in what you find. The TDPs and current ratings in the Intel specs are all maximum values. it will be useful to know if the P series always use less power than the T series, even if some T series fall within the P series current specs.

RMClock or similar get the power data from the battery which includes a chip to monitor / control power drain and charge, so that should be very accurate. While you should get consistency between this and a meter on the mains socket, I would tend to use the battery drain power method because, when running on battery, a notebook tends to shut down some of the other hardware which may affect power consumption.

One related variable is the power used by the fan. This should stay off during the idle power tests but may run at different speeds depending on which CPU and how much heat it generates.

On my Dell E6400 (P8600 + GM45 chipset), RMClock gives me the dynamic FSB option but IDA is greyed out.

John

 

TDP
Thermal design power
From Wikipedia, the free encyclopedia
(Redirected from Thermal Design Power)
Jump to: navigation, search

The thermal design power (TDP), sometimes called thermal design point, represents the maximum amount of power the cooling system in a computer is required to dissipate.
http://en.wikipedia.org/wiki/Thermal_Design_Power

 

The fan is one thing I am worried about. I have tried to take screenshots when the fan is not running for idle, and running full blast under load. I wan to unplug the fan and blow air from another source on the heatsink to eliminate this variable.

So far I have done preliminary tests on 2 T4200's, 1 T6600, and 1 P8600. Don't put a lot of faith in these numbers, but the P8600 used 1.5-2 watts less at idle than the others, and even more of a difference at load. Not exactly what I thought would happen. I want to be able to change the voltage and multiplier through rmclock so that I can compare the cores directly in addition to how they come stock.

I know. I'm seeing how much relevance the TDP value has.

 

Yeah, I was surprised also when I learned that TDP has nearly nothing to do with power consumption. I see people quote the TDP numbers all the time as if the processor would actually pull that many watts. It turns out that the Thermal Design Point is merely a measure of maximum thermal dissipation (cooling capability). Quoting TDP numbers for CPU power efficiency for a computer would be like quoting radiator size for mpg for a car.

It makes sense though. My ThinkPad has a watt gauge and I cannot do anything on the computer to get the whole system (CPU, screen, HDD, etc) to go over 25 watts. Most of the time I'm under 12 watts. It's hard to imagine a processor drawing 25 watts at normal clock speeds. Can you imagine the battery capacity (lack of)?

 

This is just a semi-educated guess. Considering the clock frequencies downclock pretty frequently even on idle, your best bet is going to try a statistical approach over a longer time period. Measure the current usage for say 4-5 minutes. You can measure while at idle, running Prime95, etc to give different numbers.

 

I've got the popcorn here Trottel.... i'm interested too
My guess is the P will have LOWER Max voltage setting so it will draw less and remain cooler under load AND why its actually labeled P.
At idle unless the P's are volted lower than the T's i don't think there will be much diff.
My P7450 is locked below 0.925V
Have a T9900 hitting the mailbox soon for comparison

 

I'm not letting the computers downclock. I could also force them to remain in a lower clocked state. Anyway, during the test there will be no switching.

 

Have you stressed your system to 100% load on CPU and GPU? I'm pretty sure it will consume more than 25W. The thing is, on battery, most people rarely stress their system, hence why the power consumption is so low. I'm guessing that at max load, the difference btw P and T processor temps are within standard deviation (btw P and T voltage ranges are typically the same).

 

I've noticed that the T series are set to 1.15v, and the P series are set to 1.1v, even while Intel says they both can be within a range.

 

I tried my kill-a-watt on my dv7 and got:

cpu and gpu idle: 27W (lcd min)
cpu and gpu idle: 32W (lcd max)
cpu 11% and gpu idle: 37W (lcd max, hd running hdtune255)

cpu load: 49W (fritz chess bench)
cpu load: 46W (wprime200)

gpu load: 69W (furmark, core1 was at 100% load)
gpu load: 65W (crysis gpubench, both cpu cores were at 0-100% load)

• The cpu differs 17W between idle and load.

• The harddrive differs 5W between idle and load.

• A rough estimate of the gpu power gives: 69-17/2-(5-1.8)-32 = 25.3 = 25W.

If we assume 25W TDP for the cpu and 35W TDP for the gpu then the cpu idles at 8W, the gpu idles at 10W and the rest of the laptop pulls 14W (decimals were discarded since i don't think the kill-a-watt or my measurements are that exact).

 

Awesome work. Very thorough as we can see how much actual power is used. It's overall system which im interested to see and very surprised a laptop doesn't use more than 100w overall. Which is less than a regular light bulb and i'd believe 3x less than a DESKTOP!?

Gosh i hate desktops. Especially P4's. Although i do love power efficient ones

 

How is the testing going Trottel?

I would be very interested in seing the kill-a-watt numbers for all cpu's on idle, on full load and doing some standard activities like watching films, surfing and gaming. Maybe even some benchmark numbers to get a performance to powerusage rating.

 

I gave my desktop (i7-920, HD4870, 22" Samsung T220) the kill-a-watt treatment and got:

Idle
laptop: 32W
desktop: 187W

Hdtune255
laptop: 37W
desktop: 189W

Fritz chess benchmark
laptop: 49W
desktop: 276W

Furmark
laptop: 69W
desktop: 290W

Furmark + Fritz chessbench
laptop: 78W
desktop: 356W

The desktop consumes quite a lot of power even at idle and using it for surfing and lighter tasks is very uneconomical compared to the laptop. If we average all benchies to get some sort of comparison number between the laptop and desktop we get:

laptop avg: 53W
desktop avg: 260W

So my desktop pulls 5 times more power than my laptop on average.

 

I think I'm going to order a kill-a-watt from Amazon today. One thing to remember with the kill-a-watt readings is that on battery power the laptop will consume a bit less. If the readings the laptop are giving me are accurate, the difference between the highest and lowest power consumption at full CPU load is about 18%, and about the same for idle. Though I can't think of a reason not to trust those results, I don't feel too confident about them. I spent several hours testing monday night and that was it. I've been pretty busy this week but I hope I can spend a couple of evenings working on this when the kill-a-watt gets here. Another consideration is that I really need to get rid of all these laptops I have laying around.

I still wish I could figure out rmclock and why those boxes are grayed out for me. I really want to be able to compare the processors at the same clock speed and voltage, in addition to their stock configuration.