Why can't we scale the frequency of a processor to a very low value?

[hshen1]

Hi,

Everyone here talks about overclocking. Obviously the thing stopping us from overclocking is the temperature.But I just feel curious why we also can't undercloking too far

What I can think of possibly are two reasons:

1. Some peripheral devices will not work when CPU frequency is too low(Is that true?)

2. We need some minimum voltage to make the CMOS circuit work. Probably the Vcc can't go below the threshold voltage of the CMOS gate. And that's the reason why we also can not underclock too far.

Anyone has answer about this?:whiste:

 

[BonzaiDuck]

Hi,

Everyone here talks about overclocking. Obviously the thing stopping us from overclocking is the temperature.But I just feel curious why we also can't undercloking too far

Actually, both temperature and voltage are limitations. Moreover, as temperature rises, more voltage is required for stability as opposed to the same clock speed with temperature held constant. It's almost a "vicious circle."

What I can think of possibly are two reasons:

1. Some peripheral devices will not work when CPU frequency is too low(Is that true?)

2. We need some minimum voltage to make the CMOS circuit work. Probably the Vcc can't go below the threshold voltage of the CMOS gate. And that's the reason why we also can not underclock too far.

Anyone has answer about this?:whiste:

Well, the Intel processors come with a ready underclocking feature. It's called "Enhanced Intel Speed Step" or "EIST." For instance, even if my processor can rev up to 4.7Ghz under load, the computer at idle settles at 1.6 Ghz and low voltage regardless of any overclock setting.

And there's also something called "sleep states."

I can underclock my system within that range, simply by turning "turbo" off in BIOS and specifying the maximum speed the processor is allowed below its default.

So . . . Maybe I misunderstood your question, but within certain parameters, it isn't as though there is some big problem with this.

 

[zir_blazer]

Actually there is a minimum operating Voltage. That is what you need to actually be able to open a transistor gate, and was around 0.6V or so, if memory servers me correctly. I recall an article in Wikipedia mentioning that, it was called Threshold Voltage. I also recall experimenting with that during 2010 with a massive undervolt and underclock of a Deneb RB-C2, and wasn't able to get under 0.6V.
http://imageshack.us/a/img694/6695/undervolt.jpg
http://imageshack.us/a/img693/7742/undervolt2.jpg (Imageshack resize seems to have killed it)
http://imageshack.us/a/img696/5256/undervolt3.jpg


Current Power Saving techniques are based on shutting down parts that you don't need instead of making them work at minimal Frequencies. I still think that a massively underclocked and undervolted Processor may be useful if you want to keep your computer on because you like to leave all your Instant Message applications, uTorrent, etc, open and ready to use. Power consumption from the Processor itself will be minimal in such state. But its the rest of the system power consumption what kill the power efficiency of doing so, not the Processor itself.

 

[BrightCandle]

I suspect you also can't go below a certain clock speed because you need a minimum ratio of clock speed to external buses to maintain basic connectivity. Threshold voltage puts a minimum voltage on the transistor switch but I suspect the lowest clock speed is still going to need more than that.

 

[know of fence]

Desktop CPUs went from 1.6 GHz to 800 Mhz on idle recenty, if I recall correctly.
The Graph shows CPU power consumption measured on the 12V rail before the voltage regulators (VRMs) for low voltage states/ Cool'n'Quiet.

[Source]http://www.lostcircuits.com/mambo//index.php?option=com_content&task=view&id=44&Itemid=1

 

[ZGR]

Can anyone get below 800 MHz on a Core i series CPU?

 

[hshen1]

Actually, both temperature and voltage are limitations. Moreover, as temperature rises, more voltage is required for stability as opposed to the same clock speed with temperature held constant. It's almost a "vicious circle."



Well, the Intel processors come with a ready underclocking feature. It's called "Enhanced Intel Speed Step" or "EIST." For instance, even if my processor can rev up to 4.7Ghz under load, the computer at idle settles at 1.6 Ghz and low voltage regardless of any overclock setting.

And there's also something called "sleep states."

I can underclock my system within that range, simply by turning "turbo" off in BIOS and specifying the maximum speed the processor is allowed below its default.

So . . . Maybe I misunderstood your question, but within certain parameters, it isn't as though there is some big problem with this.

Yes. I am quite familiar with EIST. But actually I am talking about a frequency far less than EIST supports.

 

[hshen1]

Actually there is a minimum operating Voltage. That is what you need to actually be able to open a transistor gate, and was around 0.6V or so, if memory servers me correctly. I recall an article in Wikipedia mentioning that, it was called Threshold Voltage. I also recall experimenting with that during 2010 with a massive undervolt and underclock of a Deneb RB-C2, and wasn't able to get under 0.6V.
http://imageshack.us/a/img694/6695/undervolt.jpg
http://imageshack.us/a/img693/7742/undervolt2.jpg (Imageshack resize seems to have killed it)
http://imageshack.us/a/img696/5256/undervolt3.jpg


Current Power Saving techniques are based on shutting down parts that you don't need instead of making them work at minimal Frequencies. I still think that a massively underclocked and undervolted Processor may be useful if you want to keep your computer on because you like to leave all your Instant Message applications, uTorrent, etc, open and ready to use. Power consumption from the Processor itself will be minimal in such state. But its the rest of the system power consumption what kill the power efficiency of doing so, not the Processor itself.

Your data makes sense! That's also what I think

 

[hshen1]

I suspect you also can't go below a certain clock speed because you need a minimum ratio of clock speed to external buses to maintain basic connectivity. Threshold voltage puts a minimum voltage on the transistor switch but I suspect the lowest clock speed is still going to need more than that.

Yes. That's also my concern. But I really do not really know which one(threshold voltage or peripheral) is the real bottleneck in system nowadays.:hmm:

 

[hshen1]

Desktop CPUs went from 1.6 GHz to 800 Mhz on idle recenty, if I recall correctly.
The Graph shows CPU power consumption measured on the 12V rail before the voltage regulators (VRMs) for low voltage states/ Cool'n'Quiet.

[Source]http://www.lostcircuits.com/mambo//index.php?option=com_content&task=view&id=44&Itemid=1

Hi, thanks for your information.
I am not sure why we still need 800MHz when processor is idle.But I know Intel recent processors implement very good clock and power gating for idle state on a per physical core basis. So are you sure the processor will not be clock gated during idle?

 

[nehalem256]

I think the most likely answer is that Intel and/or Motherboard vendors don't bother adding a way to induce a lower clock speed because no sane person want to

 

[Homeles]

I think the most likely answer is that Intel and/or Motherboard vendors don't bother adding a way to induce a lower clock speed because no sane person want to

No sane person wants lower idle power? Are you sure about that?

 

[KingFatty]

I would expect that the frequency can be arbitrary, but there would be a physical limitation on voltage.

But I wonder if, when you start getting pretty low on the frequency, if there are diminishing returns. So if 800 MHz is excellent, perhaps going to 4 MHz is just a waste because you are already sitting pretty at 800 and don't really save much more?

But I see no reason why a CPU couldn't just sit a zero MHz, keeping all switches from switching. Perhaps the motherboard would need to be able to accommodate that, but it just seems like a design choice and within the realm of physics/possibility.

 

[Yuriman]

I would expect that the frequency can be arbitrary, but there would be a physical limitation on voltage.

But I wonder if, when you start getting pretty low on the frequency, if there are diminishing returns. So if 800 MHz is excellent, perhaps going to 4 MHz is just a waste because you are already sitting pretty at 800 and don't really save much more?

But I see no reason why a CPU couldn't just sit a zero MHz, keeping all switches from switching. Perhaps the motherboard would need to be able to accommodate that, but it just seems like a design choice and within the realm of physics/possibility.

Isn't that basically what C-states do?

 

[Ruiner1]

I would expect that the frequency can be arbitrary, but there would be a physical limitation on voltage.

But I wonder if, when you start getting pretty low on the frequency, if there are diminishing returns. So if 800 MHz is excellent, perhaps going to 4 MHz is just a waste because you are already sitting pretty at 800 and don't really save much more?

But I see no reason why a CPU couldn't just sit a zero MHz, keeping all switches from switching. Perhaps the motherboard would need to be able to accommodate that, but it just seems like a design choice and within the realm of physics/possibility.

IME this is pretty much it. Once you have reached the lowest voltage you can work at, there is little point running below the top frequency you can manage at that level. All you do is work for longer to accomplish the same thing and spend less time completely power and clock gated.

 

[kimmel]

Circuits have a minimum voltage they will work at.
PLLs also have a minimum frequency they will lock at.

Every chip has a bounded frequency/voltage region both high and low that it will work at. Usually shown on a Shmoo plot.

http://en.wikipedia.org/wiki/Shmoo_plot

 

[ElFenix]

No sane person wants lower idle power? Are you sure about that?

intel's processors seem to use next to no power at idle. those power gate transistors really work.

 

[know of fence]

intel's processors seem to use next to no power at idle. those power gate transistors really work.

Still more than an Intel Atom or any ARM SoC at full load.

Before Haswell turned out a bit of a disappointment and a gentle kick into every enthusiasts groin. There was a lot of hype regarding the "S0ix", which basically instead of clocking down, completely send CPU and motherboard to a S3 kind of sleep instead. These changes didn't make it to the desktop though, for one reason or another. Idle power of notebooks is therefore much lower (50x), and those new sleep states even allow to recieve e-mail and stay WiFi connected. So it's not like there isn't still room for improvement, especially for chip sets and motherboards.

http://www.anandtech.com/show/6355/intels-haswell-architecture/3

 

[ghost03]

If you can control your microelectronics process parameters you can continue to scale down voltage very low, and there is significant research in that area, for wireless sensor systems which must last very long on very little battery power.

For existing x86 chips the minimum voltage is a function of parameters that are already determined by the foundry and/or chip manufacturer. Threshold voltage is more or less set by the process and can't really be changed, and when you lower voltage too far you end up in non-linear regions of operations or worse and the processor cannot continue its workflow without screwing up.

 

[ShintaiDK]

Hi, thanks for your information.
I am not sure why we still need 800MHz when processor is idle.But I know Intel recent processors implement very good clock and power gating for idle state on a per physical core basis. So are you sure the processor will not be clock gated during idle?

Why 800Mhz? Lets just imagine it only saves 1W to go to 800Mhz vs 1600Mhz. Then think on all PCs shipped and it becomes an awful large number. And you start to count the numbers of coal power plants not used. Not to mention all the money saved.

 

[know of fence]

Why 800Mhz? Lets just imagine it only saves 1W to go to 800Mhz vs 1600Mhz. Then think on all PCs shipped and it becomes an awful large number. And you start to count the numbers of coal power plants not used. Not to mention all the money saved.

The question is why 800 MHz and not lower? zir_blazer showed his 2010 Athlon running under-clocked at 100 MHz.

 

[ShintaiDK]

The question is why 800 MHz and not lower? zir_blazer showed his 2010 Athlon running under-clocked at 100 MHz.

Static power could be the reason. Other option could be program compability/reliability. Responsiveness and so on.

 

[sm625]

It doesnt really make sense for a cpu to run below 800 MHz because you hit the minimum voltage wall. So you arent gaining nearly as much compared to the "hurry up so you can enter sleep mode" design philosophy. It is much better to run the cpu reasonably fast enough to finish off whatever needs to get done in a timely fashion, and then enter a powerdown state where the clock is 0.0MHz. Once you average the time that it spends at 0MHz the average clock rate at/near idle should be much lower than even the 800MHz minimum. The problem is that windows is so badly written that the cpu isnt allowed to stay asleep for as long as it should. But at least they are finally starting to address that issue.

 

[Homeles]

Still more than an Intel Atom or any ARM SoC at full load.

Before Haswell turned out a bit of a disappointment and a gentle kick into every enthusiasts groin. There was a lot of hype regarding the "S0ix", which basically instead of clocking down, completely send CPU and motherboard to a S3 kind of sleep instead. These changes didn't make it to the desktop though, for one reason or another. Idle power of notebooks is therefore much lower (50x), and those new sleep states even allow to recieve e-mail and stay WiFi connected. So it's not like there isn't still room for improvement, especially for chip sets and motherboards.

http://www.anandtech.com/show/6355/intels-haswell-architecture/3

S0ix isn't actually supported by most Haswell chips in laptops. It's only supported in Y and U series SKUs, so basically ultrabooks and tablets. I'm rather skeptical that ultrabooks would actually make use of it, though. Connected standby is essentially useless for anything other than tablets and phones.

 

[CHADBOGA]

Those Prescott P4's were shocker's for power consumption, absolute shockers. D: