So is there some CPU underclocking/undervolting tool now?

Seems like the CPU genie died together with the whole firm, the *net is as silent as a grave regarding thsis app, while RM Clock remained frozen in time with no support for K10 or later Core 2 chips. All the underclocking/undervolting threads here are over 9 years old and mostly locked.

I know about K10stat for AMD. Is there also something which supports Core 2 and later chips I am not aware of? Or is this market totaly dead when all the later chips have automatic under/overclocking from the fab? I guess there is no need to remind that most silicon has huge headroom for future UC/OC or combination of both (as the Pentium M I just managed to run from 1.8 to 2.26 GHz on 910GL last week, while also decreasing voltage at the same time) so such SW is STILL WANTED and NEEDED.

 

depends on your targeted CPU, but the enthusiasts in this forum either use the BIOS, Throttlestop or Intel XTU

 

I might be wrong, but I'm sure if you wantto undervolt C2D you're basicially forced to pinmod it.

 

Why? It has its own P-states, just checked E4500 with RMClock, it does underclock and undervolt, but it does not allow me to change the voltage under the value for the lowest multiplier. I believe it's because of poor support for the chip, especially under 7. Or because it's HP desktop. I still remember being able to manually control some C2D CPUs under XP so at least some chips and some boards should allow for that.

Nor the Throttlestop or the XTU seems to allow for that. So am I really stuck with Pentium M or K8 and XP?

 

have u checked older versions of throttlestop? iirc previous versions also supported older CPU generations...

 

Lol Throttlestop debuted for control on Core/Core2 CPUs, and it still works marvelously thanks to @unclewebb

I have a T9500 in my T61p and it I use it to undervolt the processor while retaining max clocks and it works beautifully. Temps used to climb to 75 but now with the undervolt top out around 65

 

It does not seem to work with P-states from the screenshots, does it? Or is it only for maximum multiplier? I want to control each multiplier step.

 

When I was originally testing the newer 45nm Core 2 Duo, I found that being able to adjust the voltage for every multiplier was no longer necessary. When a CPU has a task to complete, it should enter the C0 state and get the task done as fast as possible. In terms of instructions per watt of power usage, CPUs are more efficient running at full speed compared to running at 800 MHz. As soon as a task is complete, a core should immediately stop and go into one of the low power C States. This was the logic behind the new C6 low power C State when the 45nm Core 2 Duo was introduced.

I found that holding a 45nm Core 2 Duo to a low multiplier and low voltage when idle is pointless. At idle, the cores will be spending up to 99% of their time in the low power C6 state at 0 volts and 0 MHz. For the 1% of the time when the cores are active in the C0 state processing background tasks, it really doesn't matter whether the active cores are at 800 MHz and low voltage or 2000 MHz and a higher voltage. If anything, 2000 MHz and high voltage is actually better because a fast CPU will need to spend less time working and this will allow it to spend more time idle in C6. I found that the cores are just not spending enough time at the intermediate multipliers to make any significant difference. Using SLFM mode which dropped the FSB speed in half also became useless technology after C6 was introduced. Many users are still brain washed that a slow CPU saves power but in reality, it wastes power because a slow CPU is inefficient.

https://people.eecs.berkeley.edu/~krioukov/realityCheck.pdf

All modern Intel Core i CPUs follow the same design. They save power by being able to quickly go from C6 or C7 to C0 and then immediately back to C6 or C7. In this state they are disconnected from the internal clock so they are running at 0 MHz and they are also disconnected from the main voltage rail so are at 0 volts. I do not have any Core 2 Duo C State data available but have a look at some C State data from a 4th Gen Core i.



All 4 cores are spending 99% of their time in C7. They are not even spending any measurable time in C3 or C6. The other 1% of the time is approximately 0.5% in C0 processing background tasks and 0.5% idle in C1 or C1E while transitioning between C0 and C7.

If you think that individual control of voltage for each multiplier is important, it is possible to use RM Clock with the 45nm Core 2 Duo. I found that RM Clock interfered with the CPU spending time in C6 so it ended up wasting more power than it saved but your results might be different. I heard that the developer passed away many years ago and that is why development stopped. Writing software like this is an endless money pit. That is why very few tools like this exist. In terms of time and money, costs are high and revenue is zilch. Not worth it. TS will be the next under volting tool to go bye-bye.

 

Simple look to task manager tells me that's nonsense. Even system tasks on slow CPUs - and yes, compared to modern quad-cores, some C2D with 7 is slow as it does not have 99.9 % extra power, it only has say 95 % extra power - tasks take few % of processor time constantly. Like *all* *the* *time*. Having few other usual things running and that makes it constantly at 10 % on average.

Now it always takes time for the CPU to change any state as it has to not only do that, but keep some time to stabilise. I kinda doubt that what you say is true, if the CPU was jumping from C6/7 to 0 and back each time something wanted a single instruction done, it would do nothing else than just serving the overhead of that. If you had some time chart of say 100 ms of such operation, that would tell us more.

Anyways, while all the new chips are supposed to be so wonderfull with these functions, I kinda do not see it. I can get two hours of 10+ years old battery (which should be dead by now) just by undervolting an overclocked Pentium M and having M.2 SSD in an UATA converter (instead of HDD). And it's still not really ultra-slow with Win XP as many would expect!

The reason is simple - taking down few tens of a volt is worth more than having it at so unnecessarily high voltage yet maybe in deep sleep where it consumes less - during the process. And it is of course not true that the CPU is under 0 V.

But yeah, this is the todays attitude - manufacturer gives you something which does not even reach the quality of what you had before through third-party SW, while taking the possibility of you controlling it from you. In the end the product is better from average Joe,though very slightly between generations. Comparing current CPUs to what I could do with them in the K8/Pentium M era, its so much screwed today.

 

I only wish @unclewebb implemented voltage control for the Pentium 4-M, that's a chip that REALLY needed voltage control because that thing runs way too damn hot. I'm going to show you screens from my T61p in a second here, undervolted to 1.025V max clocks during a full load TS Benchmark (vs stock voltage of 1.1375V at this clockspeed):

 

The Task Manager graph is not always a great source of information when it comes to what a CPU is really doing. Here is what it reports when running a steady 4 Thread test on a CPU with 8 available threads.



That's nonsense. The CPU is only being utilized 50% plus a hair more to process the background tasks. ThrottleStop reports that correctly but the Task Manger has grossly scaled that up. When the load is not constant, the squiggly lines are even more inflated compared to what the CPU is doing internally.

A slow CPU will have to spend more time grinding away to process all of the background tasks. Cleaning up Windows can help reduce idle power consumption.

I was not talking about the Pentium M. The low power C6 state was introduced when the 45nm Core 2 Duo was introduced. That's when ThrottleStop was written and that's when adjusting voltage for each multiplier became irrelevant. Processors are not entering and exiting C States in milliseconds. They are doing this in nanoseconds. A squiggly line task manager graph sampling the CPU every second is useless information when this is happening.

Why would a software developer spend his time working on software for a processor that is now 15 years old? That does not make any sense.

 

wait....what?! is there something i missed? are you planning to go bye-bye on us? nnnnnnooooooooooooo......

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

If that's 4 cores, that's mostly right, SMT is largely just inflanted baloon.

While processors could, in theory, do that so fast, with the clock speeds they run at, the VRM cascade never. So there are still much longer stabilisation times as that brings some serious power spikes which could render it unstable.

Plus, as you surely know: