Over-clocking with VCore offset = low power at idle

All about them.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
smilingcrow
*Lifetime Patron*
Posts: 1809
Joined: Sat Apr 24, 2004 1:45 am
Location: At Home

Over-clocking with VCore offset = low power at idle

Post by smilingcrow » Sun Nov 08, 2009 1:34 pm

I read a review at Xbitlabs of the Phenom II 965 125W and it crystallized in my mind the potential of VCore offsetting which I’d previously seen mentioned in reviews of X58 chipset boards.
In the review they included for comparison purposes the Asus P7P55D Deluxe (LGA1156, Intel P55) which supports VCore offset. It clearly shows the advantage of this setting when you compare the i5 750 versus the C2D Q9400 as the graph below shows. The i5 750 system when over-clocked from 2.66 to 4GHz only increased its idle power draw by 8W which is very good for a quad core CPU versus 25W for the Q9400 at 3.84GHz. Forget the Phenom II data as Xbitlabs used a ridiculously high VCore to raise it to 4GHz which is impractical in everyday usage. A shame they didn’t test the Phenom at a more reasonable VCore to allow a more meaningful comparison.

So is it only the Nehalem derivatives that support VCore offset? Seems a great feature for those requiring high end performance plus reasonable idle power draw especially in the more efficient LGA1156 platform.


Image

yuu
Posts: 132
Joined: Sun Aug 26, 2007 5:03 pm
Location: eu

Post by yuu » Sun Nov 08, 2009 5:16 pm

Like the DVID found on P55 GB boards

Image

my EP45-DS3 board doesn't support offset, but overclocking on default voltage keeps C1/C2/C4 states operating and voltage idles at 0.85(0.9VID) same as nehalem. 3800 @ 1.15 Load, Idle 2400 @ 0.85V

It is only the latest bios that works with this C4 enabled.

smilingcrow
*Lifetime Patron*
Posts: 1809
Joined: Sat Apr 24, 2004 1:45 am
Location: At Home

Post by smilingcrow » Mon Nov 23, 2009 12:35 pm

yuu wrote:Like the DVID found on P55 GB boards.
Interesting as I thought only Asus were taking this approach. Do you know which model Gigabyte it is?
yuu wrote:my EP45-DS3 board doesn't support offset, but overclocking on default voltage keeps C1/C2/C4 states operating and voltage idles at 0.85(0.9VID) same as nehalem. 3800 @ 1.15 Load, Idle 2400 @ 0.85V.
From my experience it’s normal for Intel platform boards to behave this way and I’ve tried this in about a dozen or more.

According to notes for recent Asus P55 BIOSs they are disabling EIST & C1E even for mild over-clocks. The note below was for the P7P55D Deluxe but appeared also for two other P55 boards that I looked at:

To quote Asus, ‘Disable EIST and CIE functions if BCLK is higher than 150MHz’ - Version 0711, 2009/09/30.

yuu
Posts: 132
Joined: Sun Aug 26, 2007 5:03 pm
Location: eu

Post by yuu » Mon Nov 23, 2009 4:39 pm

http://xbitlabs.com/articles/mainboards ... html#sect0

Probably all of them, with bios update.

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Post by Rebellious » Tue Nov 24, 2009 7:31 pm

My Phenom II 550BE with Cool'n'Quiet enabled idles at 52 Watts at the wall. When overclocked to 3.7GHz (200x18.5 at 1.425 volt) it still draws 52 watts at idle.

smilingcrow
*Lifetime Patron*
Posts: 1809
Joined: Sat Apr 24, 2004 1:45 am
Location: At Home

Post by smilingcrow » Wed Nov 25, 2009 5:08 am

Rebellious wrote:My Phenom II 550BE with Cool'n'Quiet enabled idles at 52 Watts at the wall. When overclocked to 3.7GHz (200x18.5 at 1.425 volt) it still draws 52 watts at idle.
What is the reported VCore at idle on stock settings and when over-clocked? Did you manually set the VCore in the BIOS to 1.425V?
It sounds as if you may be suffering from CnQ problems as certain recent AMD chips and/or motherboards have been showing problems in this area.
If that’s not the case then something strange is occurring as typically when you set the VCore to a fixed amount in the BIOS it doesn’t reduce even when the clock speed drops. What motherboard do you have?

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Post by Rebellious » Wed Nov 25, 2009 5:53 am

smilingcrow wrote:
Rebellious wrote:My Phenom II 550BE with Cool'n'Quiet enabled idles at 52 Watts at the wall. When overclocked to 3.7GHz (200x18.5 at 1.425 volt) it still draws 52 watts at idle.
What is the reported VCore at idle on stock settings and when over-clocked? Did you manually set the VCore in the BIOS to 1.425V?
It sounds as if you may be suffering from CnQ problems as certain recent AMD chips and/or motherboards have been showing problems in this area.
If that’s not the case then something strange is occurring as typically when you set the VCore to a fixed amount in the BIOS it doesn’t reduce even when the clock speed drops. What motherboard do you have?
Yeah, I set 1.425V in BIOS, mobo is ASUS M4A78T-E and it’s behaving as designed, no change in power consumption at idle, and a voltage-proportional wattage increase at load.

Idle power consumption is ~12W higher in Linux, but that is due to the open-source video driver. ATI did not release documentation for "Powerplay" GPU undervolting/clocking, so Linux devs are reverse-engineering* the driver.

The article you quoted from Xbitlabs looks like techno-BS. Intel is a corporate entity with a criminal record, so financing such publications is not an unusual “marketingâ€

smilingcrow
*Lifetime Patron*
Posts: 1809
Joined: Sat Apr 24, 2004 1:45 am
Location: At Home

Post by smilingcrow » Wed Nov 25, 2009 8:48 am

Rebellious wrote:Yeah, I set 1.425V in BIOS, mobo is ASUS M4A78T-E and it’s behaving as designed, no change in power consumption at idle, and a voltage-proportional wattage increase at load.
What VCore does CPU-Z report at idle at stock VCore and when over-volted?
I’ve never come across an AM2/3 or LGA775 board that works that way! Anybody else have a board with one of those sockets that behaves similarly?

It’s dawned on me that the 550 is Black Edition so maybe there’s a quirk in the BIOS that allows VCore to work that way when you over-clock using the stock buss speed. Did you over-clock purely by increasing the multi to 18.5?
Otherwise it’s a risky method to use especially so when heavily over-clocking a CPU with a low multiplier as the CPU may be unstable when running at its lowest speed setting. It may be less of an issue for AMD as their CPUs typically idle at lower clock speeds than those of Intel.
E.g. A Penium Dual-core E6300 (2.8GHz) idles at 1.6GHz so if over-clocked to 4GHz it would idle at 2.29GHz which would be a problem as it can idle as low as 0.88V using stock settings. I suppose if you disabled C3E and C4E it might help to increase Vcore at idle but then you lose the benefits of even lower power consumption.

I’d be curious if you over-clocked it by raising the buss speed whether the VCore behaved in the same way!

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Post by Rebellious » Wed Nov 25, 2009 1:02 pm

smilingcrow wrote:
Rebellious wrote:Yeah, I set 1.425V in BIOS, mobo is ASUS M4A78T-E and it’s behaving as designed, no change in power consumption at idle, and a voltage-proportional wattage increase at load.
What VCore does CPU-Z report at idle at stock VCore and when over-volted?
I’ve never come across an AM2/3 or LGA775 board that works that way! Anybody else have a board with one of those sockets that behaves similarly?

It’s dawned on me that the 550 is Black Edition so maybe there’s a quirk in the BIOS that allows VCore to work that way when you over-clock using the stock buss speed. Did you over-clock purely by increasing the multi to 18.5?
Otherwise it’s a risky method to use especially so when heavily over-clocking a CPU with a low multiplier as the CPU may be unstable when running at its lowest speed setting. It may be less of an issue for AMD as their CPUs typically idle at lower clock speeds than those of Intel.
E.g. A Penium Dual-core E6300 (2.8GHz) idles at 1.6GHz so if over-clocked to 4GHz it would idle at 2.29GHz which would be a problem as it can idle as low as 0.88V using stock settings. I suppose if you disabled C3E and C4E it might help to increase Vcore at idle but then you lose the benefits of even lower power consumption.

I’d be curious if you over-clocked it by raising the buss speed whether the VCore behaved in the same way!
Idle vcore is 0.975 both stock and overclocked. My 3.7GHz is a straight multiplier overclock 200x18.5 at 1.425 volt.

Cool'n'Quiet and C1E are both enabled, the only difference when overclocked is that CPU max p-state jumps to 3.7GHz/1.425V instead of the stock 3.1GHz/1.325V

800-1900-2400-3100 = stock
800-1900-2400-3700 = OC

My P3 kill-a-watt shows idle 52W, max ~120W stock, ~140W @3.7. Cool'n'Quiet works all the way up to 4.07 GHz (18.5x220), max temps are 42C stock, 47C at 3.7GHz, above that there’s diminishing returns.

Manabu
Posts: 21
Joined: Fri Sep 12, 2008 8:18 am
Location: Brazil

Post by Manabu » Wed Nov 25, 2009 1:18 pm

I don't know if I understood correctly this...

If CnQ is enabled, it by default always reduces the vcore to 1.025V in idle, while dropping the multiplier to the minimum possible. The intermediary steps also remain as they were. The vcores are way too conservative, so even if you do an extreme overclock in the bus speed, like 100%, this means 1.6Ghz in the lowest multiplier. That can run stable even at 0.925v in my system, let alone 1.025v.

When I undervolted phenon w/o changing the clocks or messing with k10stat, it actually was runing with like 1.2V vcore at 1.9Ghz, then drops to the 1.15v that I specified in bios for the 2.6ghz.

I'm now with an overclock+undervolt at BIOS raising the bus speed, and with k10stat I configured my clocks and voltages as following:

2990mhz - 1.2625 v ("3ghz")
2300mhz - 1.1000 v
1610mhz - 0.9375 v
1035mhz - 0.8000 v

Details in the penultimate post of the thread: viewtopic.php?t=41883&start=210

smilingcrow
*Lifetime Patron*
Posts: 1809
Joined: Sat Apr 24, 2004 1:45 am
Location: At Home

Post by smilingcrow » Wed Nov 25, 2009 1:49 pm

Rebellious wrote:Idle vcore is 0.975 both stock and overclocked. My 3.7GHz is a straight multiplier overclock 200x18.5 at 1.425 volt.
Cool'n'Quiet and C1E are both enabled, the only difference when overclocked is that CPU max p-state jumps to 3.7GHz/1.425V instead of the stock 3.1GHz/1.325V

800-1900-2400-3100 = stock
800-1900-2400-3700 = OC.
From your previous comments I deduce that the Vcore for the intermediary PStates when over-clocking is also increased at 1900 & 2400. Correct?

I’d still be curious if you over-clocked it by raising the buss speed whether the VCore behaved in the same way!
Manabu wrote:If CnQ is enabled, it by default always reduces the vcore to 1.025V in idle, while dropping the multiplier to the minimum possible. The intermediary steps also remain as they were. The vcores are way too conservative, so even if you do an extreme overclock in the bus speed, like 100%, this means 1.6Ghz in the lowest multiplier. That can run stable even at 0.925v in my system, let alone 1.025v.
It’s not going to be a problem for AMD systems as I suggested previously especially with a mild over-clock from 3.1 to 3.7.
Manabu wrote:When I undervolted phenon w/o changing the clocks or messing with k10stat, it actually was runing with like 1.2V vcore at 1.9Ghz, then drops to the 1.15v that I specified in bios for the 2.6ghz.
I wonder if 1.2V is the stock VCore for that PState?

I suppose the way to test the PState voltages would be to over-clock in the BIOS setting VCore above normal as well as the multi. Then use software to limit the multiplier and see how the Vcore compares to what it is at stock BIOS settings.

Anyway, thanks for sharing as I’d seen no reference to this before the Nehalem reviews.

Rebellious
Posts: 322
Joined: Sun Dec 09, 2007 12:53 pm
Location: EU, USA

Post by Rebellious » Wed Nov 25, 2009 2:49 pm

smilingcrow wrote:
Rebellious wrote:Idle vcore is 0.975 both stock and overclocked. My 3.7GHz is a straight multiplier overclock 200x18.5 at 1.425 volt.
Cool'n'Quiet and C1E are both enabled, the only difference when overclocked is that CPU max p-state jumps to 3.7GHz/1.425V instead of the stock 3.1GHz/1.325V

800-1900-2400-3100 = stock
800-1900-2400-3700 = OC.
From your previous comments I deduce that the Vcore for the intermediary PStates when over-clocking is also increased at 1900 & 2400. Correct?
Yeah vcore goes in steps as well, but not per core it seems.

800-1900-2400-3100 = stock
0.975-1.125-1.225-1.325

800-1900-2400-3700 = OC
0.975-1.125-1.225-1.425

Manabu
Posts: 21
Joined: Fri Sep 12, 2008 8:18 am
Location: Brazil

Post by Manabu » Mon Nov 30, 2009 5:46 pm

smilingcrow wrote:I wonder if 1.2V is the stock VCore for that PState?

I suppose the way to test the PState voltages would be to over-clock in the BIOS setting VCore above normal as well as the multi. Then use software to limit the multiplier and see how the Vcore compares to what it is at stock BIOS settings.

Anyway, thanks for sharing as I’d seen no reference to this before the Nehalem reviews.
My multiplier is locked. It seems that there is not any real hard standard. In the coments of x-bit labs review you linked they say that this behavior of enforcing C&Q in overclocking depends on the motherboard.

In AMD website, all I found was this:
http://www.amd.com/us-en/assets/content ... B_3.18.pdf
It defines the multipliers for the p-states, but only an range of aceptable voltages for each one, leaving at motherboard manufacturer discretion witch they will apply, aparently.

Klusu
Posts: 196
Joined: Thu Jul 24, 2008 10:57 am
Location: Riga

Post by Klusu » Mon Nov 30, 2009 10:50 pm

Some AMD boards do Vcore offset. But some CPUs need almost the same V at idle. One cool and quiet can not suit all. The only way is to find V for each power state for the particular CPU.

Post Reply