The "Undervolting" Guide

[TigTex]

My question is smiliar to the previous...

My laptop has an T9400 cpu (2.53ghz @ 9.5x, 2.66ghz IDA 10x). In SLFM mode this cpu goes to 800mhz (cpu-z reports it has 3x multiplier instead of 6x with half fsb).
I've been reading that this cpu in slfm has lower voltages
Core voltage in SLFM mode (V) 0.75 - 0.95
CPU-z is reporting 0.95v in SLFM, is there a way to make it 0.75v? (in slfm i am capable of getting the 8x multiplier with 0.95v) i think 0.75v is pretty stable with 6x multiplier and has a much lower power usage...

Keep up the good work with throttlestop, excelent software

[fatedquest]

My question is smiliar to the previous...

My laptop has an T9400 cpu (2.53ghz @ 9.5x, 2.66ghz IDA 10x). In SLFM mode this cpu goes to 800mhz (cpu-z reports it has 3x multiplier instead of 6x with half fsb).
I've been reading that this cpu in slfm has lower voltages
Core voltage in SLFM mode (V) 0.75 - 0.95
CPU-z is reporting 0.95v in SLFM, is there a way to make it 0.75v? (in slfm i am capable of getting the 8x multiplier with 0.95v) i think 0.75v is pretty stable with 6x multiplier and has a much lower power usage...

Keep up the good work with throttlestop, excelent software

CPU Z is unreliable to check your SLFM state, multiplier and voltage.

Use HWinfo32 instead. You can easily see when it goes into SLFM, the correct multiplier, correct voltage and even your actual RAM clock, before DDR and before Dual Channel.

[unclewebb]

ivanox1972: Core 2 thermal sensors were not designed to be accurate at reporting idle temperatures. They have several issues including the fact that many of them get stuck as the actual core temperature gets lower. There's no fix for this problem. All you can do is ignore what it says at idle. As the core temperature increases, sooner or later they all start moving but below the sticking point, they'll just sit there and keep reporting the same thing.

I'm actually booted up into Windows already.

I still don't think the 0.2V number is accurate but anything is possible. That number seems significantly below the minimum voltage number listed in the Intel docs so I'm a little suspicious.

The minimum SLFM voltage is locked at the factory in each CPU and you can't go lower than that by any method that I know of.

What I found on my T8100 is that at idle, none of this matters. According to the Windows Performance Monitor, it spends 99% of its time in C3 or a deeper sleep state so the SLFM voltage I choose in RMC or in ThrottleStop doesn't change power consumption one bit. This value is ignored 99% of the time at idle or when lightly loaded in the 45nm Core 2 mobile CPUs so I see no point in getting super technical about the perfect SLFM voltage.

TigTex: A CPU can rapidly, hundreds of times a second, be entering and exiting SLFM mode. The most accurate way that Intel provides to monitor for this results in the numbers being reported like this:

3 X 266 MHz = 798 MHz

even though internally, this is what is actually happening.

6 X 133 MHz = 798 MHz

Same total MHz either way RealTemp, ThrottleStop and CPU-Z all report this the same way now.

Core 2 CPUs have a problem that their internal timers shut down when they enter deeper sleep states. This makes calculating the bus speed accurately difficult to do without having to constantly wake the CPU up. Intel fixed this limitation in the newer Core i CPUs so even in deep sleep states the timers continue to run at their regular speed.

Constantly waking up and loading a sleeping CPU with brief bursts of full load isn't recommended so ThrottleStop avoids that. Other software might be doing this.

If the adjuster in ThrottleStop only goes as low as 0.95 volts then that is all your CPU supports. That is your minimum SLFM voltage that was read from your CPU. If you enable the SLFM option in ThrottleStop then you can use this voltage at full load without the CPU cutting the bus speed in half. It can run the SLFM voltage at whatever multiplier you set it to and the bus speed will be the normal 266 MHz. My CPU only runs reliably at 1400 or 1500 MHz when using the SLFM voltage at full load but it's a great way to make a low power chip. At idle, the actual voltage will drop even lower when it enters C3/C6, etc.

[TigTex]

Ok, looks that 0.95v is the lowest that i can get...

I have other question. How does throttlestop make cpu speed transitions?
Controlled by os?
Controlled by acpi?
Controlled by software?
because i see my cpu going from 6x slfm to 10x and 6x slfm a lot of times. acpi states are 6x slfm, 6x and 9.5x.
Using intermediate states maybe can decrease power usage no? On my pc i'm capable of running a 8x multiplier with 0.95v, better then the default 6x (normal mode) with 1.000v.

and one other question... under windows performance monitoring -> processor c states I see that cpu it's always on C0 to C2 states when on AC and goes to C3 on battery only. Is that normal?

[unclewebb]

All monitoring software has some flaky moments, even HWiNFO32 but I also prefer HWiFNO32 to CPU-Z.



ThrottleStop is designed to work at two speeds. Full speed or low speed and it ignores the OS and ACPI. You might get better power savings using RM Clock. It depends on how you use your computer. On a 45nm CPU that uses C3 at idle, I have not been able to measure any significant difference because most of these settings are ignored. At idle, my laptop consumes exactly the same amount of power whether I have it using the highest multiplier or the lowest multiplier or the highest VID or the lowest VID. All of this is ignored when in C3.

Using intermediate states maybe can decrease power usage no?

Maybe that is true but I don't have accurate enough equipment to test that. When lightly loaded, some people think it is better to run a CPU really fast so it can get the work done quickly and then when the work is finished, it will be able to spend more time in the C3 state and get better overall efficiency. Microsoft must also believe this because they changed how Windows 7 operates. Vista used to stay with the low multiplier for a bigger percentage of time but Windows 7 increases the multiplier much more quickly when lightly loaded. Use whatever method you think is best.

RM Clock still works great for many users and if you need fine control over VID and FID then I still recommend it. I will also help you get it working better with the newer 45nm CPUs and their half multipliers. For me, ThrottleStop is good enough as is. My laptop spends most of the time plugged in so saving half a watt here or there doesn't mean as much to me as it might to other users.

My Dell laptop spends most of its idle time in the C3 state whether it is plugged in or on battery power. I'm not sure what is normal. Your laptop method is probably better and might give you slightly better response when plugged in but it would be hard to notice a difference.

[TigTex]

RMclock is too old and uses too much cpu time to be eficient. Throttlestop is running for more than 4 hours now and cpu time is 0:00:00 (perfect!). Rmclock, when i watch a video, cpu usage goes to 5-10% with no reason at all.

I'm testing here right now with a powermeter.
My laptop at slfm 800mhz 0.95v uses 25watts (no load).
Also no load but at 2.53ghz 1.075v uses 25watts, 1.250v uses also 25watts...
Looks like it's useless running my cpu at lower speeds, but somehow, maybe it's just me, but i think that cpu is hotter on idle even with the same power usage (fans start spinning more times)

ps. i saw your printscreen... your C0% time is almost 0%, on my cpu, on idle it's between 30-35% on core0 70-85% on core1.
by the way: asus G71v laptop.

[unclewebb]

I'm testing here right now with a powermeter.
My laptop at slfm 800mhz 0.95v uses 25watts (no load).
Also no load but at 2.53ghz 1.075v uses 25watts, 1.250v uses also 25watts...

That's exactly what I discovered too.
VID, FID, C States, etc., doesn't make any difference at idle on the 45nm Core 2 mobile CPUs. It used to make a difference on the original 65nm CPUs like the T7500, etc. I thought maybe my watt meter was not accurate but it goes up and down just fine at idle when I adjust the screen brightness. It just doesn't do anything when I use software to adjust the CPU. Monitoring software will report that you have significantly changed your CPU but in terms of power consumption, you haven't changed anything.

If power consumption is the same at idle then core temperature will also be the same. It takes energy to create heat. Same energy, same core temperature. Maybe your fan speed is determined by the VID setting.

RMclock is too old and uses too much cpu time to be eficient. Throttlestop is running for more than 4 hours now and cpu time is 0:00:00 (perfect!).

Thanks. Someone finally noticed! When ThrottleStop is minimized to the system tray it is extremely efficient. It's also very efficient in terms of memory usage.

The C0% in ThrottleStop is based on the Intel approved method for Core i CPUs. This number is not accurate at idle on Core 2 mobile CPUs because of the issue with the timers going to sleep at idle in Core 2 Duos. As your CPU goes into a deeper sleep state, the displayed C0% number will significantly increase.

The screen shot I posted with a low C0% number at idle must have been with a desktop CPU which doesn't enter the deeper sleep states at idle on my motherboard.

[TigTex]

throttlestop is really and excelent software for laptops users and even desktop.
I think all you need now is a website with your software, no? because it's difficult to keep us updated when new versions came out and a changelog (except the one included in help files).
I'm using your software and I never faced a bug.
Keep it up

back on topic

core 2 duo T9400
2.53ghz from 1.175 to 1.075v - less 7ºC. 55ºC full load, almost 10w less power usage. Success ^^

[unclewebb]

Thanks TigTex. I agree. ThrottleStop is a useful program for many users. Updating the documentation and creating a website is on the things to do list. Hopefully some day.

Good work with your under volting project.

[fatedquest]

How do you guys check when the processor is in a C1, 2, 3 or deeper sleep state? I can't find anything that indicates that from Windows Performance monitor or resource monitor.