Overclocking And Overvolting


kingoddball said:
@ Mali:

A SLIDER?

Is there a slider to over clock? WTF!? AWESOME!

Its not a physical slider, just one in a program. That would be sweet, though.
 
Last edited by a moderator:
kingoddball said:
@ Mali:

A SLIDER?

Is there a slider to over clock? WTF!? AWESOME!
Yeah, that's a hell of a great device, isn't it :) The N900 users had to beg for it and the community had to accomplish it. Pandora does it out of the box :)
 
Last edited by a moderator:
I thought the battery meter was supposed to be pretty good?
Can't we use that?
 
The battery meter calculates based on momentary power consumption afaik, so it's not suitable as a benchmark.
 
What spec do the Droid and N900 run at? OPP5? I've seen posts about Droid running stable at 1Ghz, and others have taken it to 1.3Ghz, but I don't imagine it was terribly stable.

Also curious about the heat of the cpu when operating at such high voltage and frequency.
 
mali said:
The battery meter calculates based on momentary power consumption afaik, so it's not suitable as a benchmark.
Is things like quake and n64 full screen only?
 
Last edited by a moderator:
Na-Noo said:
mali said:
The battery meter calculates based on momentary power consumption afaik, so it's not suitable as a benchmark.
Is things like quake and n64 full screen only?

There's no reason for them to be. It's a regular linux desktop - should be able to run windowed...but why do you ask?
 
Last edited by a moderator:
quadomatic said:
What spec do the Droid and N900 run at? OPP5? I've seen posts about Droid running stable at 1Ghz, and others have taken it to 1.3Ghz, but I don't imagine it was terribly stable.

Also curious about the heat of the cpu when operating at such high voltage and frequency.
OPP5, I believe N900 has voltage scaling.
 
Last edited by a moderator:
quadomatic said:
Na-Noo said:
mali said:
The battery meter calculates based on momentary power consumption afaik, so it's not suitable as a benchmark.
Is things like quake and n64 full screen only?

There's no reason for them to be. It's a regular linux desktop - should be able to run windowed...but why do you ask?
Run timedemo and see what the battery meter is reporting, I remember Craig saying it was very accurate as they were using a special chip.
 
Last edited by a moderator:
Na-Noo said:
Is things like quake and n64 full screen only?
The reason for the battery meter being of no use as a battery life benchmark is, that it's too inaccurate. Also running quake and benchmark battery life wouldn't work well either, because it doesn't show what battery life in an every day scenario would be like. You would need a scenario with variable CPU load to cover the full spectrum. If you overclock and run Quake, it will use the additional power to render more frames and therefore use more power. In an office/web-surfing situation you would have periods of idle and some spikes of higher CPU load in between.
 
Last edited by a moderator:
I believe that it is accurate for what it is used. What I mean is that you have to charge a battery fully and then let it run out of juice while you do certain tasks till the device shuts itself off. Do the same at a different clock speed and compare both results. A battery meter is no benchmarking tool, no matter how accurate it is as it calculates the remaining battery life based on momentary power consumption.

edit:
To make it more clear and example: You read an eBook and it shows 16h. You run n64 and it shows 7h. You verify that the battery meter is accurate, cause you had a stop watch running at the same time. Now you mix both scenarios and do another test. The battery meter will always be accurate at the time of measurement, but it won't give you the exact battery life in the end.
 
I was thinking worse case scenario.
It would tell you that if you was overclocking to run a demanding app, you would loose x amount of battery time, x percentage more battery usage would occur.

In general usage, well it should also be able to tell you how much less/more battery life your getting.

Trouble is everyone's usage will vary, but if you ran say a timedemo, that is something that would be consistent for everyone.

For me, that would be enough. Think maybe we have different requirements?
 
mali said:
The battery meter calculates based on momentary power consumption afaik, so it's not suitable as a benchmark.

That's true, it estimates based on the immediate consumption, like any battery meter that I know of. However MWeston did show us some tests and pictures based on variable usage, where the previously given estimates proved to be accurate. Certainly around the 10 hour mark, I believe.
 
Last edited by a moderator:
Na-Noo said:
I was thinking worse case scenario.
It would tell you that if you was overclocking to run a demanding app, you would loose x amount of battery time, x percentage more battery usage would occur.

In general usage, well it should also be able to tell you how much less/more battery life your getting.

Trouble is everyone's usage will vary, but if you ran say a timedemo, that is something that would be consistent for everyone.

For me, that would be enough. Think maybe we have different requirements?
That's all true. The problem is that you can't generalize the timedemo for other scenarios. A higher clock speed that lowers battery life in Quake might raise battery life in an office/web-surfing scenario.

Regarding accuracy of the battery life meter software, I don't know how it is programmed, so I speculate too much. For example I use Battery Bar on my 1810t. It shows 2:01 left right now, while Vista's software says 3:20. Battery Bar is relatively accurate, because it saves a history of previous usage scenarios and mixes those results with actual measurements. Maybe Pandora offers this too, dunno.
 
Last edited by a moderator:
Is voltage scaling in a later kernel, I can't remember.
I know you need a custom one to change the OPP, but that's different than scaling.
 
craigix said:
The battery meter should always be accurate to about a minute. This isn't some shitty old GP2X tech.

Stunning, are you using radioactive atomic decay of the battery as an indicator or something cause that's mighty accurate. Hell I don't even boil eggs to within 1 minutes accuracy!
 
Last edited by a moderator:
With OPP5 you'll probably see alot of 1ghz+ Pandora overclocks. I'm not sure what the OPP3-5 levels even stand for, on the N900 overclocking voltage is chosen by a number from 0 to 72, with 0 being 0.6v and each increment is a 0.0125v increase of ontop of that. So say you used a voltage setting of 24 for 250mhz, it would be 0.6v + 0.3v for a final voltage result of 0.900v

N900 for 600mhz uses a stock setting of 60 (1.35v) and downscales the voltage depending on the frequency - 500mhz uses 48 (1.20v), 250mhz uses 38 (1.075v)etc.

I've played very thoroughly with overclocking on my N900 and tweaked every single frequency to the lowest voltage increment possible. Currently for everyday use I have it set to a maximum frequency of 700mhz with a value of 36 (1.05v) for voltage in hopes of increasing battery life. It doesn't really seem to make a hell of a difference though, I still get just shy of 2 days worth of use seemingly regardless of what frequency I use. Likely the same on the Pandora, overclocking should have very little effect on battery life.. being smart with the screen brightness and disabling wifi/bluetooth radios when you don't need it should make a more significant difference.

It will do 1100mhz at Nokia's stock voltage value for 600mhz but not a single increment lower. 1150mhz, the highest frequency the custom nokia kernel I'm using will allow (artificial cap by creator so people don't potentially break their N900s) is stable with a value of 63. The higher frequencies scale worsen in efficiency (as typically happens in PC overclocking), about 3 increments to the voltage per 50mhz or so. If this path stays in the same linear fashion then my N900 should be capable of hitting a peak of 1300-1350mhz stable with the maximum voltage setting of 72. Could 72 be OPP6? No idea.

My N900 overclocking results seem to be very good in comparison to other people's results on the N900 forums, but the N900 overclocking thread is very full of "noise" from people just copy-pasting other people's settings and FUD, only a handful of people are actually setting and testing the voltages themselves so we don't have a very good sample of how the yields really are for overclocking and what voltages they may require.
 
Back
Top