C1E - enhanced halt actually makes power consumption worse

[taltamir]

I tested the feature that lowers the intel multiplier to 6x while not in heavy demand, and found it actually increased power consumption!

If I remember correctly, I read in anandtech that the phenoms loose about 5-10% performance when their dynamic power is enabled, and that the Core2 loose about 2-5%.
I figured, eh, it will be a good first step to improving performance, before I even try to OC, and then it will probably OC better too.

Well, I recently decided to test exactly what difference it makes. The results astounded me.
Kill-o-watt measurements with it off were 105 watts idle with a spike to 106, 107, or 108 every few seconds, almost always 107 (well, idle except for vista SLOWLY filling up my ram).
With it on, it was also 105, but spiking to 115 watts every few seconds.

With OCCT maxing out my ram and CPU at the same time, both settings showed a solid 179watts. (which they should, since both should be running at 9x multiplier solid).

I did have some stability issues with it on. The solution was to stop undervolting (I overclock my Q6600 to 3ghz and underclock it slightly), without the underclock it increased my min and max by 3 watts.

Overall it surprised me how USELESS this "feature" is, I expected it to be about 10watts lower and possibly worthwhile for its performance loss, but it actually made the electricity usage go up, both due to the need for higher voltage, and also due to spikes being higher. I think those spikes might coincide with the CPU going from 6 to 9x ALL the time... I saw that on both me E8400 and on the Q6600, every few seconds, even when nothing is running, it will go to 9x multi and then drop back down.

 

[Idontcare]

Your results are obviously going to depend on what kinds, and how much, of crapware you have loaded on your system that is idle most of the time but periodically is calling home to see what's up.

When my systems drop to 6x multi they stay there for 10's of minutes, yes like watching paint dry I've sat there watching my kill-a-watt for minutes as you have done.

On my QX6700 I get a pretty solid/consistent 5-6W decrease in power consumption at the wall when my idle system goes from 10x to 6x multi (it is a QX6700). Clearly everyones results will vary.

 

[dug777]

I use Speedstep and the Asus EPU-6 engine thingo with my Q6600 and P5Q PRO, and the EPU-6 tool actively controls vcore depending on load, so in the lowest setting (which the machine is almost always in in office use/browsing/email etc) my vcore is .992 (when multi is 6x), under full load my vcore is 1.104 (multi is 9x).

Undervolted of course (bios is 1.1875 I think, vid is 1.325 from memory), which also makes a big diff to my load especially (15-20W iirc). Didn't change my idle much (maybe 3W), since the EPU-6 thing dynamically dropped the idle voltage anyway.

Never tested speedstep off, but I don't notice my multi flickering like you describe, watching it closely in CPU-Z.

EDIT: C1E and EIST are different things, I think, but damned if I can find out clearly what is what and what what is what

 

[Idontcare]

Originally posted by: dug777
Never tested speedstep off, but I don't notice my multi flickering like you describe, watching it closely in CPU-Z.

What are you guys running in the background that is doing this to your C1E?

I've even got Norton 2009 installed and my multi doesn't do this.

C1E does what it is advertised to do, so clearly you guys have something loaded in the background on your comps that is periodically spiking the CPU activity just enough to trigger C1E to ramp your multi.

Perhaps it is the windows disk indexer? Or Diskeeper invisitasker? Something is going on for you, but faulting C1E as simply being dysfunctional is probably the trivial but incorrect diagnosis here.

 

[myocardia]

Originally posted by: Idontcare

Originally posted by: dug777
Never tested speedstep off, but I don't notice my multi flickering like you describe, watching it closely in CPU-Z.

What are you guys running in the background that is doing this to your C1E?

Dug said his multi doesn't fluctuate. Mine doen't either, BTW.

 

[taltamir]

Originally posted by: dug777


I use Speedstep and the Asus EPU-6 engine thingo with my Q6600 and P5Q PRO, and the EPU-6 tool actively controls vcore depending on load, so in the lowest setting (which the machine is almost always in in office use/browsing/email etc) my vcore is .992 (when multi is 6x), under full load my vcore is 1.104 (multi is 9x).

Undervolted of course (bios is 1.1875 I think, vid is 1.325 from memory), which also makes a big diff to my load especially (15-20W iirc). Didn't change my idle much (maybe 3W), since the EPU-6 thing dynamically dropped the idle voltage anyway.

Never tested speedstep off, but I don't notice my multi flickering like you describe, watching it closely in CPU-Z.

EDIT: C1E and EIST are different things, I think, but damned if I can find out clearly what is what and what what is what

C1E = lower multiplier and vcore.
EIST = Allow a windows driver/programs to manipulate core speed manually (aka, gigabytes dynamic energy saver sends commands to the bios that EIST uses to modify the curernt clockspeed).
I just read more about EIST, I wonder if I should have enabled it as well as C1E, rather then instead of....

The EPU is a chip that manages your voltages etc, it also disables power circuitry that isn't being actively used at the time...

@Idontcare:
1. Only thing happening was vista supercache was caching (I didn't wait for it to finish, cause frankly, it took FOREVER!, it was loading about 2MB a second, and I have 4GB of ram)
2. I idled at the same 105 W on both of them. If the 6x multi was showing decreasing consumption, it was less then 1W and rounded away by KILL-A-WATT
3. I had increase voltage back to stock to make it stable at the new setting, and that actually increased my idle and max power consumption by 3 watts, negating any possible benefits.

 

[Idontcare]

Originally posted by: myocardia

Originally posted by: Idontcare

Originally posted by: dug777
Never tested speedstep off, but I don't notice my multi flickering like you describe, watching it closely in CPU-Z.

What are you guys running in the background that is doing this to your C1E?

Dug said his multi doesn't fluctuate. Mine doen't either, BTW.

Well I'll be! Smack me with a wet fish then...I fail at reading.

Sorry dug for my pointless post.

 

[myocardia]

Originally posted by: Idontcare
Well I'll be! Smack me with a wet fish then...I fail at reading.

Hah, at least you don't do that as often as I do.

Originally posted by: taltamir
3. I had increase voltage back to stock to make it stable at the new setting, and that actually increased my idle and max power consumption by 3 watts, negating any possible benefits.

There's the reason your power consumption didn't go down. C1E and SpeedStep are meant to be used by OEM's and non-undervolters. I'm sure it isn't quite as efficient as a well-undervolted CPU (done correctly, in other words). It's just meant to be used by lazy people.

 

[taltamir]

Actually, power consumption did not go down when it was at the same exact voltage, but it was unstable... when i increase voltage to make it stable it went UP.

1. undervolted system: 105 watt idle. with spikes to 107. Load : 179watt
2. undervolted system with C1E on: 105 watt idle and unstable. with spikes to 115. Load: 179 watts
3. stock voltage system with C1E on: 108 watt idle with spikes to 124... load: 183 watts.


Although... i wonder how it will handle a specific load that is about 50% cpu load on the 6x... maybe that would be lower on the 6x multi then it is on the 9x... but considering the 9x it will take less percentage (same amount of calculations, but out of 150% total).

 

[dug777]

Originally posted by: myocardia

Originally posted by: Idontcare
Well I'll be! Smack me with a wet fish then...I fail at reading.

Hah, at least you don't do that as often as I do.

Originally posted by: taltamir
3. I had increase voltage back to stock to make it stable at the new setting, and that actually increased my idle and max power consumption by 3 watts, negating any possible benefits.

There's the reason your power consumption didn't go down. C1E and SpeedStep are meant to be used by OEM's and non-undervolters. I'm sure it isn't quite as efficient as a well-undervolted CPU (done correctly, in other words). It's just meant to be used by lazy people.

I disagree entirely

As per my first post, the combination of undervolting, Speedstep and the Asus EPU thing means that my idle and low cpu usage state (which comprises the vast majority of my computer usage with a Quad and Office/FF/mail/tunes & non hi-def media) means that my low power vcore is .992v (while my multi is 6x).

With the best will in the world I doubt .992 will be stable at 9x, my VID is 1.325v, and 1.104v at full load (9x multi) is p95 64-bit stable. (all voltages from CPU-Z, Bios is either 1.175 or 1.1825 (or something around that mark, I forget right now.)

The combination of these things allows me such a low idle and low load power usage.

 

[taltamir]

well.... have you tried undervolting manually?

 

[dug777]

Originally posted by: taltamir
well.... have you tried undervolting manually?

I apologise for the extreme lack of clarity in my posting

I undervolt manually in addition to leaving all 'Speedstep' items enabled (C1E and EIST), and I also use the Asus EPU hardware/software.

So I manually change the BIOS vcore from auto to whatever I previously described (I don't know what it uses as 'auto', but my VID from CPU-Z is 1.325, from memory).

 

[Idontcare]

Originally posted by: dug777
So I manually change the BIOS vcore from auto to whatever I previously described (I don't know what it uses as 'auto', but my VID from CPU-Z is 1.325, from memory).

Doesn't doing that then disable the voltage reduction function of C1E when the system is idle and it drops the multi down to 6x?

It does for me. If I specify a voltage at any value in the BIOS other than AUTO then I still get the multi reduction with C1E but C1E will no longer automatically decrease the Vcore any further than it's standard idle value (which I manipulated by way of setting it in the BIOS).

 

[dug777]

Originally posted by: Idontcare

Originally posted by: dug777
So I manually change the BIOS vcore from auto to whatever I previously described (I don't know what it uses as 'auto', but my VID from CPU-Z is 1.325, from memory).

Doesn't doing that then disable the voltage reduction function of C1E when the system is idle and it drops the multi down to 6x?

It does for me. If I specify a voltage at any value in the BIOS other than AUTO then I still get the multi reduction with C1E but C1E will no longer automatically decrease the Vcore any further than it's standard idle value (which I manipulated by way of setting it in the BIOS).

Couldn't tell you I'm afraid

It's complicated by the ASUS EPU, which has control over my voltages as well.

With the arrangement as described, I can watch the vcore and mutliplier (and FSB -EPU does that too!) change (in CPU-Z) as the EPU program moves through its load settings.

The difference between undervolting and not undervolting (with everything enabled as described) is primarily load as I mentioned earlier (idle doesn't change much), but whether C1E and the EPU hardware/software work towards the same idle/low load state voltage (which seems logical) or not, i couldn't tell you.

So confusing

 

[taltamir]

yes, it seems to start going beyond our ability to test / figure out what the heck is going on. Cause I have no clue what the EPU does (the the similar thingie I have in my board, which is gigabyte)

 

[myocardia]

Originally posted by: Idontcare
Doesn't doing that then disable the voltage reduction function of C1E when the system is idle and it drops the multi down to 6x?

Yes, it does. Intel won't let them call it SpeedStep, if it doesn't, I hear.

 

[dug777]

Originally posted by: myocardia

Originally posted by: Idontcare
Doesn't doing that then disable the voltage reduction function of C1E when the system is idle and it drops the multi down to 6x?

Yes, it does. Intel won't let them call it SpeedStep, if it doesn't, I hear.

So my vcore drop at x6 and low load is all due to the EPU?

 

[taltamir]

from wikipedia I read that EISP is speedstep. which is reducing the multiplier on demand by software / driver / os, but NOT reducing the voltage. Voltage reduction is supposed to be the C1E (which is the CPU reducing its own speed and voltage by its own volition, not based on what a driver / software / OS tells it to do).

Although it makes sense for C1E to not reduce voltage if it is already below VID.