1Ghz Liquid

[Guest Redline66]

Dear cookers, now that the Eclair kernel sources are available, can we hope to get soon a rom for our Liquid with CPU max speed increased to 1Ghz instead of the 768 Mhz of the stock device?

[Guest jayziac]

Where is the Eclair kernel source for the liquid? It's not on Acer's website...

BTW if they are really available, we'd get better performance by getting Froyo or JIT working instead of just overclocking.

Edited May 28, 2010 by jayziac

[Guest drigerott]

.......................................... :(

[Guest xian1243]

*sigh* The Eclair kernel sources are NOT out! Please stop making useless threads, and stop spreading disinformation. We have DONUT'S source, NOT ECLAIR'S.

[Guest wytek]

oh nice is it out?!?!? link thx :( :(

[Guest Nesli]

Where is the Eclair kernel source for the liquid? It's not on Acer's website...

BTW if they are really available, we'd get better performance by getting Froyo or JIT working instead of just overclocking.

JIT provides acceleration only for applications, which are compilled in non native code. So nothing like a 3D games cannot be accelerated using JIT..OC is good and usefull anyway;)...

[Guest jayziac]

I personally don't care much about overclock. I like Acer's decision to undervolt and underclock. 768 Mhz is fast enough to run compiled code anyway, the current interpreted Java in Android is what slows things down mainly. If it's possible I would undervolt & underclock it down to 512 Mhz for most everyday stuff (assuming JIT works) to conserve battery power. JIT gives performance gains of 200% - 500% whereas overclocking from 768 Mhz to 1 Ghz only 33% and uses up more battery.

[Guest lgcmn]

as i can see our 768 Mhz is equal to 1 Ghz (tested only in 3d games))

no sense in overclocking only less battery

:(

[Guest Auxx]

Acer did only underclock Scorpion CPU (Snapdragon is a SOC, not CPU), 3D from AMD is not touched. 1ghz/768mhz are only peak values. For most of your interactions CPU runs on like 300mhz and goes less then 250mhz for idling. You will not improve battery life by slowing down. You only can improve it by high quality software, which is aware of ARM architecture. Linux kernel is not yet in the club, btw. But Google did a lot in that direction.

[Guest demolition23]

I have done xian recomedetion and my liquid runs @ 576 mhz, its very smooth as it was without setcpu and battery life has improved a lot.

[Guest Auxx]

I have done xian recomedetion and my liquid runs @ 576 mhz, its very smooth as it was without setcpu and battery life has improved a lot.

That's because of software errors. Sometimes I quit browser, Dalvik goes insane and starts to eat 99% of CPU time until I reboot. This makes phone hot + battery depletes like insane. Of course lowering CPU speed will make it deplet slower in such ocasions. But slowing down CPU is not a good solution. Fixing software is.

[Guest jayziac]

I think slowing CPU clock only works if the core voltage is lowered too (requires kernel modifications). Acer did lower the voltage by .1 from default 1 Ghz Snapdragon. There may be additional savings if lowered further since Power ~= Voltage^2

[Guest Quipeace]

Nope, lowering clocks decreases power consumption, per second (not per cycle), so lowering clockspeed does increase battery life.

by the way, its P=U*I, not P=U^2.

[Guest gamero_m]

check this out!!! phh has posted linpack results and acer liquid appears as 998.4 MHZ , so he made it?

http://www.greenecomputing.com/apps/linpac...pack-by-device/

[Guest Andrea1Liquid]

yes, but because they are using the donut's kernel by disc0.

[Guest phhusson]

I think slowing CPU clock only works if the core voltage is lowered too (requires kernel modifications). Acer did lower the voltage by .1 from default 1 Ghz Snapdragon. There may be additional savings if lowered further since Power ~= Voltage^2

Nope, lowering clocks decreases power consumption, per second (not per cycle), so lowering clockspeed does increase battery life.

by the way, its P=U*I, not P=U^2.

He said ~=Voltage^2 not =, which is quite correct if R is supposed constant.

R doesn't change much depending on the frequency, but R is higher as the cpu is warmer, so I'd say that for a cpu, P~= Voltage^2.5 or 3.

Anyway that's not the question.

Lowering clocks decrease power consumption when not idling.

So it decreases power consumption, if you continuously run something eating CPU.

Now, the thing is that for a fixed operation (let's say linpack benchmark, but it applies to anything, it could be launching MMS apps or whatever) the number of operations is constant.

So when lowering clock, you decrease a bit power consumption (*0.5 in frequency gets circa *0.8 in instant power consumption), but you make it take even longer (*2 in this example), so for the total computation, you consumed *0.8*2=*1.6 => MORE than at higher frequency !

Now, to link those two last facts, we need to consider how much we need to overvolt the cpu to (over/under)clock it.

This is quite exponentional.

So, that's when there is something to do by the cpu.

When there is nothing to do, you have to know that the "normal" clock doesn't apply to the CPU.

Unlike some people seem to believe here, when idling, the cpu doesn't run at 768MHz or even 250MHz

It runs at 19.2MHz in current kernel configuration, it can go to 32kHz, and you can even make it completely stop. (I don't know wether I'll include this option or not, it seems the N1 doesn't use it, so I don't think I'll try it.), so clocks when idling are mostly irrelevant.

So to sum up... Stop saying the result in battery life is obvious, it's everything but obvious.

Edit: Argh, stupid acer. I've just seen that they don't ramp down clock on WFI, so the 19.2MHz is wrong on most kernels (but mine :(

Edited May 30, 2010 by phhusson

[Guest CedricGatay]

Edit: Argh, stupid acer. I've just seen that they don't ramp down clock on WFI, so the 19.2MHz is wrong on most kernels (but mine

It explains battery drainage when wifi is activated :(

BTW you've done a wonderful job :(

[Guest drigerott]

one thing... not all Liquid support 1ghz..... the maximum that no one had problem is 920 mhz

[Guest Quipeace]

He said ~=Voltage^2 not =, which is quite correct if R is supposed constant.

R doesn't change much depending on the frequency, but R is higher as the cpu is warmer, so I'd say that for a cpu, P~= Voltage^2.5 or 3.

Anyway that's not the question.

Lowering clocks decrease power consumption when not idling.

So it decreases power consumption, if you continuously run something eating CPU.

Now, the thing is that for a fixed operation (let's say linpack benchmark, but it applies to anything, it could be launching MMS apps or whatever) the number of operations is constant.

So when lowering clock, you decrease a bit power consumption (*0.5 in frequency gets circa *0.8 in instant power consumption), but you make it take even longer (*2 in this example), so for the total computation, you consumed *0.8*2=*1.6 => MORE than at higher frequency !

Now, to link those two last facts, we need to consider how much we need to overvolt the cpu to (over/under)clock it.

This is quite exponentional.

So, that's when there is something to do by the cpu.

When there is nothing to do, you have to know that the "normal" clock doesn't apply to the CPU.

Unlike some people seem to believe here, when idling, the cpu doesn't run at 768MHz or even 250MHz

It runs at 19.2MHz in current kernel configuration, it can go to 32kHz, and you can even make it completely stop. (I don't know wether I'll include this option or not, it seems the N1 doesn't use it, so I don't think I'll try it.), so clocks when idling are mostly irrelevant.

So to sum up... Stop saying the result in battery life is obvious, it's everything but obvious.

Edit: Argh, stupid acer. I've just seen that they don't ramp down clock on WFI, so the 19.2MHz is wrong on most kernels (but mine :(

Its very obvious on notebook/desktop systems, you try running prime95 at 2 and 4GHz, total power consumption decreases even though prime95 has to work longer to get to the same point. (Yes I actually measured that, and totla "profit" depends on how well the CPU clocks down after the process) I just assumed it was the same with smartphones :o

P(F V)=cFV^2

c = constant

F = frequency

V = voltage

I can see where P ~= V^2 comes from, but it would probably be clearer if it was written like P = (...)V^2, I thought he meant another formula.

What frequency does the Liquid currently run when idling?

PS: The drivers are written in C/C++ right? Might have a workaround for the multitouch bug if I understand the current situation correctly.

EDIT: Typo, and once again, Thanks for the time you guys are putting into this :(

@below, was that sarcastic? XD

Edited May 30, 2010 by Quipeace

[Guest drigerott]

EDIT: Typo, and once again, Thanks for the time you guys are putting into this :(

yes, thanks very much.. :(

[Guest phhusson]

Its very obvious on notebook/desktop systems, you try running prime95 at 2 and 4GHz, total power consumption decreases even though prime95 has to work longer to get to the same point. (Yes I actually measured that, and totla "profit" depends on how well the CPU clocks down after the process) I just assumed it was the same with smartphones :(

Just do that with decent values. And I don't think you run your cpu at 2 and 4Ghz with the same voltage.

Like frequencies on which the CPU is meant to run at.

What frequency does the Liquid currently run when idling?

It's a bit messy, but I think it's 128MHz. (It does WFI though, so it's not exactly like running at 128MHz)

[Guest Quipeace]

Just do that with decent values. And I don't think you run your cpu at 2 and 4Ghz with the same voltage.

Like frequencies on which the CPU is meant to run at.

It's a bit messy, but I think it's 128MHz. (It does WFI though, so it's not exactly like running at 128MHz)

Was just trying to illustrate my point :(. actual values were 2.3GHz and 3.2GHz on an i7 860.

WFI is a very low power mode right?

Edited June 1, 2010 by Quipeace

[Guest phhusson]

Was just trying to illustrate my point :(. actual values were 2.3GHz and 3.2GHz on an i7 860.

Then you're not. You're stating it's obvious, with only two values... (well one since you do only comparaisons)

WFI is a very low power mode right?

No, think about it as an acpi C1 cpu mode.

[Guest Quipeace]

Then you're not. You're stating it's obvious, with only two values... (well one since you do only comparaisons)

No, think about it as an acpi C1 cpu mode.

1. Lets leave it alone alright :(, we're wasting forum space XD

2. Got it, halt mode :(