Intel Speedstep Problem On Some Motherboards?

[WR304]

EDIT: See my next post down

--------------------Original Post-------------------------------------
Something I hadn't realised is that Intel Speedstep appears to work differently depending what motherboard you have.

I recently got a Gigabyte P35 DS4 rev1.0 motherboard. It has an Intel E6700 Core 2 Duo CPU fitted.

Intel speedstep works fine at stock speed 10*266, 2.66ghz. It drops voltage and the multiplier when the CPU is idling to 6*266 1.6ghz as expected.

As soon as you change the FSB (Front Side Bus) to a higher number in the BIOS it stops working though and the CPU is stuck at full speed all the time.

Apparently it's the same on the Asus Striker Extreme motherboard too where any overclock disables Intel speedstep stright away.

With the other Intel Socket 775 motherboards I have here (Asus P5K and Asus P5W DH Deluxe) it works fine: You can increase the FSB and the multiplier will still drop down to 6x at idle.

It's why you always seem to have to actually use these things before finding out about their quirks.

Is it a widespread issue that most Intel Socket 775 motherboards disable speedstep when the FSB is increased?

[tehfire]

Yes, EIST not working when overclocked is quite a common problem. I wonder what the actual cause is, but I've heard about it for a long time now.

I wonder if there's a way to know before buying a board...

[angelkiller]

But isn't this a safety feature? Lets say 266x10 is stable at 1.3v. (stock volts) Then EIST lowers the multiplier and voltage to 6 and 1.1v respectively and hypothetically. EIST lowers multiplier and voltage to a static value, whatever that may be.

Suppose you then OC your E6700 to 350x10. What happens when EIST reduces you CPU to 350x6 at 1.1v? My point is that when you up the FSB, the faster you CPU will operate at the reduced voltage. This could cause instability depending on what FSB you use or how low EIST reduces voltages.

I can't say this is true for sure, but it makes sense....

[Lawrence Lee]

I assume it's a safety feature as well.

EIST doesn't disable on the Gigabyte DS3 if you change FSB though, only if you change vCore.

[smilingcrow]

Intel speedstep works fine at stock speed 10*266, 2.66ghz. It drops voltage and the multiplier when the CPU is idling to 6*266 1.6ghz as expected.

As soon as you change the FSB (Front Side Bus) to a higher number in the BIOS it stops working though and the CPU is stuck at full speed all the time.

I’ve never heard of this happening before although certain boards will disable VCore from changing once you raise the FSB but still allow the clock speed to change. In these cases the VCore is typically raised as well even if left on standard in the BIOS.
I documented this issue in a number of threads in this sub-forum last year including some workarounds. If you search this sub-forum for threads I started last year you’ll find the details.

In your case is the VCore still variable or is that fixed also? If the VCore is fixed then losing the ability to change clock speed isn’t such a big deal anyway.

Note. Sometimes in these situations CPU-Z or other utilities don’t show accurate VCore readings so you have to verify the voltages by other means; measure the power draw or CPU temperature.

Yes, EIST not working when overclocked is quite a common problem. I wonder what the actual cause is, but I've heard about it for a long time now.

In my experience of working with about 10 C2D boards only one of them showed this issue although I’m not sure what the current situation is with P35 boards. My G33 board doesn’t have this problem.

I wonder if there's a way to know before buying a board...

Not really unless someone wants to start a thread here for people to post details on their boards. Even that is not foolproof as different BIOS versions can display different results. I first noticed this issue with I think it was an Asus P5B Deluxe. The initial BIOS I used was fine but on upgrading to a newer version this issue reared its ugly head.

But isn't this a safety feature?

I’m sure that’s why they implement it and also the same reason they boost the VCore and fix it above the stock maximum value when you raise the FSB and leave VCore on standard.

Suppose you then OC your E6700 to 350x10. What happens when EIST reduces you CPU to 350x6 at 1.1v? My point is that when you up the FSB, the faster you CPU will operate at the reduced voltage. This could cause instability depending on what FSB you use or how low EIST reduces voltages.

I agree but it’s only a problem in certain situations. In the example above you would typically have to raise VCore in the BIOS to achieve 3.5GHz and in that case the VCore is fixed at that value regardless of what Speedstep does to the clock speed.
I’ve never come across a board that allows you to raise VCore in the BIOS and still allows Speedstep to modulate VCore with clock speed. The only way around this I suspect is to pin mod the CPU for a higher VID.

This problem is more commonly found in C2Ds with very low multipliers e.g. the E6300 has a multi of 7 so if it was stable at 3GHz at stock voltage it would idle at 2.57GHz which would be too high for the lowest Speedstep voltage setting. With an E6600 in the same scenario it would idle at 2GHz which should be fine.
This problem could become more common as Intel is moving to a FSB of 333 for most CPUs which will require lower multipliers for the same clock speed. Penryn will change the dynamic as well as it may run higher at stock voltage but we don’t know yet how well it under-volts.

There are ways around this issue. If you have a C2D with a multi of 7 and you are overclocking it then Speedstep doesn’t offer a significant benefit so you can disable it. Otherwise I suggest this method:
Disable Speedstep and C1E in the BIOS and set the Multi to 6; it will then boot at 6 x FSB at stock voltage so will be stable. Use RMClock to set Speedstep settings that are stable and to re-enable C1E and Speedstep e.g. Multi 6 = 1.225V, 7 = 1.25V, 8 = 1.275V etc. This way you can have your cake and eat it.

[austinbike]

So the short answer is that when I run my system in a way that was not intended by the manufacturer, it doesn't work like the manufacturer intended it to?

If you are running it out of spec, you should start with the assumption that it work work right and be delighted if it does instead.

[smilingcrow]

So the short answer is that when I run my system in a way that was not intended by the manufacturer, it doesn't work like the manufacturer intended it to?

If you are running it out of spec, you should start with the assumption that it work work right and be delighted if it does instead.

Are you a professional pessimistic by any chance? The evidence is that most motherboards work fine and that this is issue only affects a small number of them.

A way around this issue is to raise the FSB using clockgen as then Speedstep will continue to work provided that VCore is left at stock setting.

[WR304]

I just tried Speedstep again on the Gigabyte P35 DS4, whilst watching the frequencies in RMClock, and the multiplier seems to be working now!

The multiplier will drop to 6x, even when the FSB is increased, but the vcore seems to be staying at 1.325v.

Ignore the CPU temps as the readings are taken from two different PCs at different clock speeds.


Speedstep on Asus P5K with Overclocked Intel E6700 CPU 10*300, 3ghz


Speedstep on Gigabyte P35 DS4 with overclocked Intel E6700 CPU 10*333, 3.33ghz

[smilingcrow]

I just tried Speedstep again on the Gigabyte P35 DS4 watching the frequencies in RMClock and the multiplier seems to be working now!

The multiplier will drop to 6x even when the FSB is increased but the vcore seems to be staying at 1.325v.

What is VCore set to in the BIOS?

[WR304]

Vcore is Auto in the BIOS on both computers.

It does mention in the Gigabyte BIOS help that it may automatically adjust vcore upwards when the CPU is overclocked though.


CPU settings E6700 on Gigabyte P35 DS4 motherboard

[smilingcrow]

Vcore is Auto in the BIOS on both computers.

Does is have the option to set it to Normal as opposed to Auto? I think with some BIOSs Auto means it will over-volt if you raise the FSB whereas Normal leaves it at stock settings. I’m going to check my Gigabyte BIOS now.

[WR304]

I tried changing the Gigabyte P35 DS4 voltage settings from Auto to Normal.

It can drop the voltage down now: there's a small blip on the pic below but it still stays at full Vcore most of the time.

Although it doesn't show it clearly on the RMClock graphs something significant must have changed though.

Compared to the same motherboard and CPU on Auto settings the Normal setting has dropped idle CPU temps by 10c from 47c down to 37c!


Speedstep on Gigabyte P35 DS4 with "Normal" Voltage settings

[smilingcrow]

I tried changing the Gigabyte P35 DS4 voltage settings from Auto to Normal.

I just did the same and the difference in power draw is 9W at idle and 25W running Orthos (large FFT) @ 2.43GHz.
I wonder if the difference between Manual and Auto is widely appreciated!
Just to make it clear to get the lower VCore you need to set the following:

System Voltage Control = Manual
CPU Voltage Control = Normal

To get the higher VCore set System Voltage Control = Auto

[crabnebula]

I believe what is happening is that when selecting the "Normal" setting, the BIOS is usig the VID table to adjust voltage. On auto, it is probably overriding the VID table entirely, which is why it will drop the multiplier, but not the voltage.

You're lucky to have such a setting on the Gigabyte. I've just built a new PC based on an Asus P5K-E and there is no "Normal" setting for the Vcore. As soon as you raise the FSB, Vcore becomes fixed, regardless of what the VID table says. Even using RMClock has no effect. This is very annoying because you can't overclock the FSB and use EIST/RMClock effectively. Only the multiplier can be lowered based on load, which has only a small effect on temps/watts.

If anyone is aware of a workaround I'd be glad to hear about it.

[crabnebula]

Actually I've found a workaround now. I boot at the CPU's native FSB and raise the FSB with SetFSB afterwards. By doing this, EIST will scale both Vcore and multiplier based on load, even once the FSB is overclocked.

[smilingcrow]

Actually I've found a workaround now. I boot at the CPU's native FSB and raise the FSB with SetFSB afterwards. By doing this, EIST will scale both Vcore and multiplier based on load, even once the FSB is overclocked.

I reported on the same workaround last year but using Clockgen instead – see here.
Are there any other utilities that allow FSB changes in software? The more the merrier.

[crabnebula]

Indeed you did report this.

I don't know of any other programs. SetFSB seems to have the widest mobo support.

The drawback with this workaround is that the FSB goes back to its BIOS default if you standy/resume.

[smilingcrow]

The drawback with this workaround is that the FSB goes back to its BIOS default if you standy/resume.

Is this typical behaviour when using software control of FSB?

The Asrock Conroe945G-DVI behaves like this even when you over-clock in the BIOS which surprised me.

[Flandry]

The drawback with this workaround is that the FSB goes back to its BIOS default if you standy/resume.

Meaning that you have to reboot to get dynamic FSB again?