Sign in with
Sign up | Sign in
Your question

Speedstep on Core Duo 2

Tags:
Last response: in CPUs
Share
September 21, 2007 10:13:30 AM

I just got myself an Intel CPU, having changed from Athlon64. The latter had very good tools, by means of using "minimal power settings" to make the CPU run at lower clock speeds during idle periods. I can't see how to achieve this with the Core Duo 2. CPUz always reports it running at full speed, so does System Info.
Help much apreciated!

More about : speedstep core duo

a b à CPUs
September 21, 2007 10:28:33 AM

go to BIOS and enable SPEED STEP
September 21, 2007 10:29:28 AM

That should be default on the mobo. Don't know about yours let me check.
Related resources
a b à CPUs
September 21, 2007 10:32:46 AM

yes thats strange SPEED STEP is ENABLED for many motherboards , mine was enabled too
September 21, 2007 10:34:55 AM

It looks like you might need a BIOS update. This may help


AnandTech.com [ Article: abit AB9 QuadGT: Test Lab Update]

Quote:
Follow up Impressions

While the numbers of problems reported seem vast, it has been our experience that most P965 motherboards had similar amounts of issues upon release. Most of these have been solved through BIOS updates, driver releases, or revised board designs. abit has already solved a number of problems with the last two beta BIOS releases and are well on their way to solving the final issues that also include getting EIST, TM2, and C1E operating correctly again in the next production release BIOS.
September 21, 2007 10:40:23 AM

EIST is enable in BIOS, and it is the latest BIOS version (v21)...
September 21, 2007 10:45:51 AM

I could be wrong, but I don't think windows has any control over EIST. Nothing on the Abit forum?
September 21, 2007 10:58:05 AM

It appears that the EIST problem was fixed in this version. I don't know what else to tell you.

http://www.abit.com.tw/page/en/motherboard/motherboard_detail.php?pMODEL_NAME=AB9+QuadGT&fMTYPE=LGA775&pPRODINFO=BIOS
BIOS ID:11
1. Enhanced overclocking ability.
2. Modify the Voltage adjusting function.
3. Fixed an issue that the system hangs during Vista installation, when using a Kentsfield CPU.
4. Fixed an issue where the system sometimes hangs on POST code 2A.
5. Fixed the JMB363 performance issue.
6. JMB363 now works as intended when enabling ICH8R RAID ROM.
7. The "EIST adjusting" function now works as intended.
8. BIOS compiled date: 3/27/2007


September 21, 2007 11:00:00 AM

I've now played around with the power settings under Vista a little, and changed it from "Balanced" to "Power saver". Then clock speeds dropped from 3000 to around 2000, which is still way more than I need/want, no?
September 21, 2007 11:05:35 AM

Check CPU-Z for the multiplier, I'm not sure about yours but the Q6600 drops from 9 to 6. As a matter of fact, my 3G OC drops to 2G. so your probably good.
September 21, 2007 11:12:05 AM

Apparently, I was in left field.

Enhanced Intel SpeedStep® Technology How To Document

Quote:
4. Finally Enhanced Intel SpeedStep technology must be turned on in the OS. Currently, for Windows* XP SP2 operating systems Enhanced Intel SpeedStep technology by default is off. To turn it on do the following:

* Under Control Panel %u2013 open Power Options
o Under the Power Schemes pull down menu
+ To turn Enhanced Intel SpeedStep technology ON select, "Minimal Power Management", power scheme.
+ To turn Enhanced Intel SpeedStep technology OFF select, "Always On", power scheme.

To verify that Enhanced Intel SpeedStep technology is enabled do the following:


I never use always on, because I shut down the monitor and the HD after about 30 min. I never use Hibernate or suspend though, too many problems with it.
September 21, 2007 11:30:11 AM

I'm actually using Vista, so the System screen looks different. But even though CPUz reports 2000MHz, System reports 3.06Ghz. But it doesn't get updated there, so it might be a spike of CPU activity from loading the screen?
Multiplier is indeed 6.
September 21, 2007 11:34:17 AM

I would rely on CPU-Z. I've seen, and I can't remember exactly where, the OS reports one number and CPU-Z reports another. The CPU-Z number increases as you load the system.
September 21, 2007 11:38:23 AM

You can run Prime95 25.4 with CPU-Z open and see the Multi change. Don't run it very long without a temp sensor running, or you could make the CPU quite hot if you don't have proper cooling.
a b à CPUs
September 21, 2007 11:48:29 AM

How about using the power button?
You can't get any greener than shutting the machine off!...lol

Look, if you need it on 24/7 then don't worry about it. If not shut it down when you're away....screwing with it, you'll wind up having to re-install or worse mucking up your system !!!!!!!!!

You can always set up your power options to shut down all the system components when you want.

Hope this helps!
September 21, 2007 11:56:43 AM

Did you read the thread? He already solved the problem.
September 21, 2007 12:08:17 PM

Weird, on Vista with balanced scheme, EIST works fine over here on a C2D E4300, no need to use Power Saver.
a c 96 à CPUs
September 21, 2007 12:11:03 PM

Naujoks said:
I've now played around with the power settings under Vista a little, and changed it from "Balanced" to "Power saver". Then clock speeds dropped from 3000 to around 2000, which is still way more than I need/want, no?


EIST works by reducing the multiplier to 6 in all Pentium M/Core-based chips. (The multiplier for Pentium 4s/Ds varies- it is 12 in Pentium 4-Ms and 14 in 800 MHz FSB Pentium 4s and Ds that have EIST.) Your chip has a 333 MHz FSB clock rate to yield its 1333 MHz effective FSB rate, so the 6x multiplier used in EIST puts the idle speed at 2000 MHz. So you are in fact seeing EIST work properly.

If you want a lower idle clock speed on an Intel chip, you have to use a CPU with a lower FSB speed. Here are the idle clock speeds of the various Intel Core 2-based desktop chips:

800 MHz FSB (Celeron 4x0, Pentium Dual Core, Core 2 Duo E4x00): 1.20 GHz
1066 MHz FSB (Core 2 Duo E6x00 and X6800, Core 2 Quad Q6x00 and QX6700, Xeon 5100/5130 < 2.00 GHz): 1.60 GHz
1333 MHz FSB (Core 2 Duo E6x50, Core 2 Quad QX6850, Xeon 5100/5300 2.33 GHz+): 2.00 GHz

This 6x multiplier as the base EIST multiplier was set in the days of the original Pentium M Banias with a 400 MHz FSB, resulting in a 600 MHz idle speed. That worked well, and the 533 MHz FSB bump for the Pentium M Dothan 2.0 resulted in an 800 MHz idle speed, which also was decent. The Core 2s hail back from the Pentium M laptop CPU lineage, so the 6x multiplier for EIST stuck with them. It is decent on 800 MHz FSB chips but start to be a little less than effective as the FSB speeds shot upwards. In fact, some CPUs don't even have EIST as their top multiplier is 6 (e.g. 2.00 GHz Xeon.) Intel should have incorporated the same FSB down-clocking used in the GM965/PM965 laptop chipset into the 30-series desktop chipsets. The 965 mobile chipset reduces the FSB clock from the base 200 MHz (800 FSB) to 133 MHz (533 FSB) when the chip is idle, dropping the idle speed from 1.20 GHz to a more battery-friendly 800 MHz. This would have let Intel set a nice low idle speed at the same fixed 6x multiplier along all of their desktop chips- the chipset would always drop the FSB to the same speed regardless of what the maximum FSB speed was. This works well on the laptops, so I can't see why it wouldn't work on the desktop.

Oh, well, I suppose Intel will do what they always do- use the "bigger hammer" approach (process node scaling -> lower Vcore across the board) rather than a more elegant approach (FSB scaling) to solve their problems. I just wonder what they could do if they used *both* approaches...
September 21, 2007 12:23:19 PM

Thanks for the additional information. It would be a problem with Notebooks, but for a desktop I'll take the 2G. My Northwood 3.4 never stepped down at all and I let it cook anyway. It's only a heat issue for me, and with my Ultra 120 it's no big deal.
September 21, 2007 12:38:40 PM

Wow MU_Engineer, that was a comprehensive answer. Thanks for that!
Well, I'm glad all's working as well as it should, but it's a shame that things are still running at such high speeds when all I do is surf the net, write an email or download p... files.
September 21, 2007 12:47:11 PM

As MU_Engineer alluded to, if your that concerned then lower your FSB to 1066. If you start doing something that you need the extra clocks for, then go into BIOS and raise it back up to 1333. I don't really think it's that big of a deal though, it's only 400Mhz. Let it eat.
September 21, 2007 1:11:10 PM

Well, let's face it, convenience DOES come into play with this at some stage.
I understand there are programs that lower voltage etc, but i think they involve a lot of tinkering. One-click solutions would be great. Like "office use", "dvd" and "gaming". A real software niche! Anyone?

Zorg said:
As MU_Engineer alluded to, if your that concerned then lower your FSB to 1066. If you start doing something that you need the extra clocks for, then go into BIOS and raise it back up to 1333. I don't really think it's that big of a deal though, it's only 400Mhz. Let it eat.

September 21, 2007 2:08:08 PM

I don't even understand what you just said. Just let it eat.
September 21, 2007 2:19:14 PM

I'm saying that I can't be bothered to go into the BIOS everytime. And I'm lamenting the fact that there is not easier way to keep my system cool and quiet during those many hours when I'm not using high performance apps. Over and out.

Zorg said:
I don't even understand what you just said. Just let it eat.

September 21, 2007 3:00:21 PM

Well 2G is relatively cool and quiet. Most people that do any serious OCing turn it off anyway, for stability reasons.
September 21, 2007 3:36:17 PM

Far from being and OC person, I am a cool'n quiet person. QUIET, more than anything. My CPU fan's running at 900rpm. When it's on, it's so quiet I can even hear it at all when in bed.
September 22, 2007 12:57:53 AM

MU_Engineer said:

Oh, well, I suppose Intel will do what they always do- use the "bigger hammer" approach (process node scaling -> lower Vcore across the board) rather than a more elegant approach (FSB scaling) to solve their problems. I just wonder what they could do if they used *both* approaches...


MU, great post. However, Intel has announced bus-geyserville (it was announced in prior IDFs). Unfortunately, I can't find anything on intel.com about it. This might be in Merom E-step + Santa Rosa (I can't remember). Wikipedia states this:

"Socket P processors are capable of throttling the FSB anywhere between 400-800 MHz as necessary."

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

You are correct about Bearlake/Crestline differences. Desktop chipsets and mobos are cheaper, and don't even have DPSLP# and DPRSLP# routed.

The real power savings come from sleep states, and not speed step. Part of the reason why Merom and Penryn have a lower TDP than their desktop counterparts is that there is a high residency in DeepC4 and C6 sleep states, which do things like flush the L2 cache, allowing the CPU to go to an even lower voltage. Desktop CPUs have a maximum of C1. "C1E" means that the CPU enters C1 sleep state, and goes into LFM (low frequency mode). When the CPU is in C1, it has no clock. However, LFM is beneficial because (1) the FSB consumes less power, and (2) the CPU is actually able to wake up faster than if it were NOT in LFM. This has to do with the voltage ramp and PLL locking. Sorry if this is boring and off-topic, but hopefully someone finds it interesting. :ange:  Also, please note that this post does not necessarily reflect the views of Intel Corp.

a c 96 à CPUs
September 22, 2007 2:33:02 AM

Quote:
MU, great post. However, Intel has announced bus-geyserville (it was announced in prior IDFs). Unfortunately, I can't find anything on intel.com about it. This might be in Merom E-step + Santa Rosa (I can't remember). Wikipedia states this:

"Socket P processors are capable of throttling the FSB anywhere between 400-800 MHz as necessary."


Interesting. After looking through the Intel tech docs for the Core 2 Duo mobile processors on the GM/PM965 chipset, it seems that 533 seems to be the minimum FSB speed used (as 800 MHz is the lowest LFM mode listed.) If it is capable of 400 MHz operation, I am curious why they did not enable it on the LV and ULV CPUs and drop the idle speed down that much more.

Quote:
The real power savings come from sleep states, and not speed step. Part of the reason why Merom and Penryn have a lower TDP than their desktop counterparts is that there is a high residency in DeepC4 and C6 sleep states, which do things like flush the L2 cache, allowing the CPU to go to an even lower voltage.


Is this ever true. My laptop doesn't see the C3/C4 sleep state when it is plugged in- the BIOS hides it and it can only go into C2. When I suspend-to-RAM with the laptop plugged in and then resume with it not plugged in, the CPU only sees the C2 state as the highest C-state and battery life suffers by about 20%, not to mention running 5-8 C warmer at idle. C-states make quite a bit of difference in battery life and thermal dissipation, that is for sure.

Quote:
Desktop CPUs have a maximum of C1. "C1E" means that the CPU enters C1 sleep state, and goes into LFM (low frequency mode). When the CPU is in C1, it has no clock. However, LFM is beneficial because (1) the FSB consumes less power, and (2) the CPU is actually able to wake up faster than if it were NOT in LFM. This has to do with the voltage ramp and PLL locking. Sorry if this is boring and off-topic, but hopefully someone finds it interesting.


I certainly do. I wonder why desktop CPUs have only C1 sleep state when the higher sleep states really do help save power and heat while idling- and most computers spend 99% of their time idling. This is especially baffling when almost all of the metrics used today in servers are power and thermal-related. Putting C3 or C4 sleep state in there would do wonders for idle power consumption, which even with many servers is their typical load. Perhaps number-crunchers and HPC setups care more about reducing maximum dissipation by five watts rather than idle dissipation by five watts, but everybody else would benefit more from the opposite arrangement.

Quote:
:ange:  Also, please note that this post does not necessarily reflect the views of Intel Corp.


Ah ha, we have somebody who has firsthand knowledge of these topics. Glad to have you here.
September 22, 2007 4:15:14 AM

I assume that these sleep states relate to suspend and hibernate. I have never had very good luck with those low power settings in any MS OS. I haven't tried Vista because I won't use it.
a c 96 à CPUs
September 22, 2007 5:05:02 AM

Zorg said:
I assume that these sleep states relate to suspend and hibernate. I have never had very good luck with those low power settings in any MS OS. I haven't tried Vista because I won't use it.


No, they are lower-power states for the processor. They do things like turn off chunks of unneeded cache when the chip is idle. These should work no matter the OS, as long as the OS supports CPU C-states. Vista supports the C-states in CPUs (haven't used it to verify) and I know that Windows XP does as well as Linux.
a c 96 à CPUs
September 22, 2007 1:50:55 PM

Zorg said:
So these additional sleep states are only for the mobile CPUs?

Edit: I looked and it appears that it is only on the Penryn mobile CPUs.

Techgage - Intel Penryn and Nehalem New Details


C-states have been in mobile CPUs for some time. My 5-year-old Mobile Pentium 4-M had C1 and C2 states and the Core 2 Duo mobile chips have C1-C4 states. The deep-power-down (C6) state appears to be new to Penryn though.
September 22, 2007 7:17:05 PM

Zorg said:
So these additional sleep states are only for the mobile CPUs?

Edit: I looked and it appears that it is only on the Penryn mobile CPUs.

Techgage - Intel Penryn and Nehalem New Details


Great link. C6 is a big deal and a big challenge. You can see in one of those slides that Vcc is basically 0V. This effectively wipes away every bit of state maintained in a processor. The processor has to do something similar to an OS hibernate to disk in order to save and restore.

Also note that the author of the article seems to think that it takes seconds to enter a C state. This is not true. The CPU can enter and leave C1 in a matter of bus clocks. C3/C4 is in the order of microseconds. Under a typical workload, your mobile part is able to enter and exit a C state in between keystrokes of a word processor, and definitely in-between frames of watching a DVD (it was designed this way).

Deep sleep states would have value in most desktop applications, but not so much in a data center. I don't think Google's servers ever idle, for instance. It would be a waste of their money (aren't they hurting? :D  ) to have unused computational capacity. This is just my take, though, and I could be wrong about this.

Also, mobile-on-desktop is catching on. For instance, I'm writing this from my beautiful new 24" iMac with an E-step Merom and a Crestline chipset. The sucker is cool, and quiet (you might say cool 'n quiet).

Time to go to the Texas game. MU_Eng, doesn't the Big XII stink this year?
September 22, 2007 8:00:46 PM

MU_Engineer said:
C-states have been in mobile CPUs for some time. My 5-year-old Mobile Pentium 4-M had C1 and C2 states and the Core 2 Duo mobile chips have C1-C4 states. The deep-power-down (C6) state appears to be new to Penryn though.
Yeah, I'm aware of that. I meant only mobile, which appears to be the case. Not that I'm all that concerned, I almost never use my notebook anyway.
September 22, 2007 8:02:42 PM

Phanboy said:
Great link. C6 is a big deal and a big challenge. You can see in one of those slides that Vcc is basically 0V. This effectively wipes away every bit of state maintained in a processor. The processor has to do something similar to an OS hibernate to disk in order to save and restore.
I hope it works better than the MS OS version, of course they may have it worked out in Vista.
!