Change your old memory, batterylife benefits! (56k warning)

[jiopsi]

Ok, my T40 takes, with all powersavers on:

CPU 1.5GHz Banias@ (0,716v 6x)
MR 7500@ 60/60MHz
LCD@ dimmest
WLAN@ OFF
DVD/ports@ OFF

, about 9W. There are three steps that Im currently investigating and going after to taking this consumption even lower.

1. Already ordered Samsung 4GB SSD which should save up to 1.3W-2.0W to my older 5K80 drive.
2. Change 2100 3B WLAN, not so big, but have to do it to gain 54Mbs speed. (investigating)
3. Was surprised about how much old DDR consumes power. Here is table of my current 512MB Mosel and newer alternatives difference:

EDIT 4.12: the change old 512MB to newer 512MB saves about 0.4-1.0W.

All these changes should bring power consumption to about 6W level. Which should give batterylife with my 95Wh battery (6600mAh+2200mAh ultrabay) up to 16h where the memory change would mean huge 2.5h difference.

Any opinions, suggestions and better powerconsumption wise product finds are welcomed.

[aaa]

Wow, didn't think of old ram affecting things...

A Dothan cpu would make a teeny difference, considering I was able to find my 1.5 Dothan for a few dollars, it might be worth it.

For wireless, I'm thinking the 2200bg would be good power wise, dunno about the 2915. Never liked the Intel drivers though, it's why everybody hates the Intel cards.

Oh, linky. Broadcom cards use less power? (edit, nevermind, at idle they use way more).

[jiopsi]

For wireless, I'm thinking the 2200bg would be good power wise, dunno about the 2915. Never liked the Intel drivers though, it's why everybody hates the Intel cards.

Oh, linky.

Superb find, yeah 2200BG seems pretty good choice.

As for CPU, in my memory, although smaller process 130nm-90nm the Dothan still uses bit more power at 6x, because of the bigger 2MB cache which cannot be shut down. How low can you underclock your 1.5GHz Dothan?

[aaa]

For some reason it's stuck in my mind that the Dothan uses less power in spite of the 2M cache. Dunno why exactly though.

For you, it's a teeny difference (from .732 to .7), but I don't always lock my cpu at 600mhz. My Dothan is capable of 1ghz/.732v, and the higher speeds are way lower as well. So it makes a big difference if you let the cpu ramp up on demand.

[jiopsi]

For you, it's a teeny difference (from .732 to .7), but I don't always lock my cpu at 600mhz. My Dothan is capable of 1ghz/.732v, and the higher speeds are way lower as well. So it makes a big difference if you let the cpu ramp up on demand.

You can use 0.7v in your Dothan, passed NHC test? Yeah in higher clocks, Dothan is definitely better, because L2 share of total power consumption gets smaller. I can get my 1.5 Banias pass NHC test at 1.084V 15x, how about yours?

EDIT: just tested, that I can pass that test also at 0.7v 6x, older results where with older bios etc.

[aaa]

15x = .908v

Dothans always are able to do 600/.7, it's unusual for a Banias to go that low. Presumably they can do lower but .7v it's the lowest available.

[jiopsi]

Ok, Im on the fence with Dothan. But just bought 2200BG with 20€, should have it tomorrow, and new DDR memory. Here is the table which upgrades are comming, I will keep updating when they arrive.

[sojourner]

jiopsi, really inpressed by your post. The findings regarding power consumption are eye opening indeed. Especially liked the last chart (very readable/understandable).

Was curious, what software are you using to capture these readings?

[RealBlackStuff]

That looks like a photo of an Excel spreadsheet.
There's a very good, free, screen capture program (WinSnap) available here:
http://www.ntwind.com/software/winsnap/ ... rsion.html
Highly recommended.

[rocky01]

... considering I was able to find my 1.5 Dothan for a few dollars ...

Now you know that's not nice, enticing casual reader like that, without details and all

Do tell, and perhaps advice for finding speedier T40 RAM upgrades maybe on the cheap and best moves for speeding up entire system with faster (but energy efficient?) hard drive upgrade. I recall reading on the subject in these same pages but wasn't yet mtivated at the time.

Replies as it suits you (anyone) as it's my favorite thinkpad subject! I'll search the archives but site's search engine is not that smart, at least with me running it

[ricerocket]

6w? Good Lord. I need to read up on undervolting. I'm at 15w with the brightness cranked down and with the wireless(es) off.

[gunston]

interesting

[jiopsi]

Changed my WLAN to faster 2200 BG, couldn't notice accurately any power consumption changes, so I have just to trust those specs.

Applied AS5 thermal paste at the same time, and that gave very good results. http://forum.thinkpads.com/viewtopic.ph ... 730#358730

New memory and SSD should arrive next week.

[aaa]

... considering I was able to find my 1.5 Dothan for a few dollars ...

Now you know that's not nice, enticing casual reader like that, without details and all

Do tell, and perhaps advice for finding speedier T40 RAM upgrades maybe on the cheap and best moves for speeding up entire system with faster (but energy efficient?) hard drive upgrade. I recall reading on the subject in these same pages but wasn't yet mtivated at the time.

Replies as it suits you (anyone) as it's my favorite thinkpad subject! I'll search the archives but site's search engine is not that smart, at least with me running it

Well someone put it up on ebay starting at .99, and no one else bid. Really, the lower speed chips like that are not in demand, while the chips that are just a fraction faster (like 1.7) are going for $30.

Ram, on the T40 you're stuck at pc2100 speed no matter what you do. I suppose you can find lower latency (CL2 vs the standard CL2.5), but it's probably not worth the hassle. I'm planning to switch my 2x512 for a single 1gb, since 1 stick usually uses less power, on top of the newer-using-less-power factor. Of course, if you really love your battery life above all, you can settle for 256m or 128m , as reducing the amount makes the most difference.

For HDs, 7200rpm is usually the way to go, as there is no noticeable power difference (it is louder though). Only way to significantly reduce power on the HD front (aside from spinning it down) is to get an SSD.

[aaa]

Some numbers I dug up. I think Kingston's power numbers are not particularly accurate, because they can't possibly be using the same chips all the time, so it's probably overstated a bit. Anyhow, ram power estimating involves alot of guesswork and depends on how many times the OS acceses the ram, so I used the "low usage" values in Micron's power calculator to get my "Total Estimate".

Code: Select all

Kingston 1GB PC2100 (KVR266X64SC25/1G):
Total Estimate: n/a
IDD0 Estimate: 3.9w

Kingston 512MB PC2100 (KVR266X64SC25/512):
Total Estimate: n/a
IDD0 Estimate: 2.9w

Micron 1GB - Grade -40B, 2.5v
Total Estimate: 3.2w
IDD0 Estimate: 3.2w

Micron 512MB - Grade -40B, 2.5v
Total Estimate: 2.8w
IDD0 Estimate: 2.78w

Micron 256MB -Grade -40B, 2.5v
Total Estimate: 2.2w
IDD0 Estimate: 1.55w

Micron 256MB -Grade -265, 2.5v
Total Estimate: 1.5w
IDD0 Estimate: 1.15w 

So apparently going from 2x512 to 1x1gb is supposed to save 1.9-2.4w, an awful lot. And the grade of the chips seems to make a 20-30% difference for those 256 modules. Too bad I couldn't find the matching data for the bigger modules.
Links:
Kingston Ram Datasheets
Micron DDR SODIMM List
Micron Ram Power Calculator

[rocky01]

O.K. I guess I'll have to look for a 1GB stick as well.

I just need to zero in on best 7200RPM drive option and read up on how to make sure to image the hidden recovery partition again. It seems the HD is clearly biggest bang for buck. Then I think it's time to look at the chips out there. Now that I'm officially out of warranty, I'm ready to futz.

Appreciate the reply!

[jiopsi]

Made FSB400 Dothan 2MB 90nm/Banias 1MB 130nm comparison chart about power consumption from Intel datasheets. X is voltage and Y is power. Power consumption should be between TDP and Max.AutoHalt depending on load %. You have to use little averaging eye to smooth those bumps.

You can see that in same voltages, Dothan with 2MB L2 cache clearly sucks more power in idle. But it takes it back with lower voltages and faster process time with same multiplier.

Could you aaa tell your 1.5GHz Dothan's lowest possible voltages in stress with all multipliers, so we could compare it my 1.5GHz Banias.

[FTC]

Jiopsi,

Those graphs are very interesting, but I think they are taken at standard processor voltage and do not take into account the fact that dothans can be undervolted much more than Banias, so if you drawed those lines at average reachable voltage per each frequency you would see how easily dothans do make for the difference in power consumption (given that P=I*V) and even take advantage by their own. Also note that dothans are faster than banias clock per clock, due to the fact that they have more cache, and this ought to be put in the equation also..

[jiopsi]

I know that FTC, and explained that in the post:

You can see that in same voltages, Dothan with 2MB L2 cache clearly sucks more power in idle. But it takes it back with lower voltages and faster process time with same multiplier.

, and that is why I need voltages with any multipliers from aaa or somebody other with 1.5GHz Dothan FSB400.

[aaa]

Dump of the settings I use:

Code: Select all

static struct cpufreq_frequency_table dothan_1500[] =
{
        OP( 600,  700,  700,  700,  700),
        OP( 900,  700,  700,  700,  700),
        OP(1000,  732,  732,  732,  732),
        OP(1100,  812,  812,  812,  812),
        OP(1200,  844,  844,  844,  844),
        OP(1300,  876,  876,  876,  876),
        OP(1400,  892,  892,  892,  892),
        OP(1500,  908,  908,  908,  908),
        { .frequency = CPUFREQ_TABLE_END }
};

[jiopsi]

Have you aaa used any stress programs or do you have any random freezes with your settings?

[aaa]

Have you aaa used any stress programs or do you have any random freezes with your settings?

Nope, I've been using these settings without problems for a few months.

[jiopsi]

Ok, here are my results, 6x 10x 15x tested with SuperPI 2MB and others NHC internal tester:

6x 0.700 4:58 (freezes randomly, 0.716 without)
9x 0.796
10x 0.844 3:27
11x 0.872
12x 0.924
13x 0.972
14x 1.036
15x 1.084 2:40

[aaa]

I'm remembering that mine weren't absolute minimums, I only heavily tested 6x, 9x, 10x, and 15x. The rest were overshot so I wouldn't have to test, because testing takes so long.

[jiopsi]

-----

[jiopsi]

Made some very ruff chart. Without speed factor on same multiplier, on TDP side, Banias uses clearly more power at any multiplier. On the idle side Dothan wins and ties 11x~, but calculated from the trend, Banias wins ~10x.


But I think this has assured me to change to Dothan, because Banias advantage area is so small, and with speed factor even smaller or maybe zero.

[jiopsi]

Just bought Dothan 725 1.6GHz from ebay with 27€, so I will be testing power consumption difference also with CPU change.

[gunston]

Dothan 745 M - 1.8Ghz

6x 0.7v
8x 0.732v
10 .796v
12x .876v
14x .956v
16x 1.052v
18x 1.148v

Tested with Prime95, working absolutely error free!!!

[jiopsi]

Made some preliminery tests where to compare component changes:


Dimmest, all off (WLAN/LAN/Ports/DVD)
1. Idle, pretty sure every component were in idle.
2. SuperPI to test CPU in stress.
3. Winamp to test CPU near idle mode.
4. Ram Performance to test worst case scenario in memory, 2,2W difference between idle and read/write is pretty close what specsheet shows. And if at IDD4/5 ram change is near at what specs predicts, then you can trust them at idle mode also.
5. HD Tach to test HDD in read mode.

[Nicola]

Ok, my T40 takes, with all powersavers on:

CPU 1.5GHz Banias@ (0,716v 6x)
MR 7500@ 60/60MHz
LCD@ dimmest
WLAN@ OFF
DVD/ports@ OFF

, about 9W.

What is 9W?