Low power C States and long term reliability

GeneO · 01 Mar 2012

Hi,

The Intel volume 1 data sheet for second generation core Desktop processors states the following about C-states:

"Caution: Long term reliability cannot be assured unless all the Low Power Idle States are enabled"

I would like to hear what people think about that.

I just want to turn them all off and use EIST so I can control the lowest multiplier from Windows Power Management. I would enable C1E but that affects SSD performance. And the lower C3 and C6 give lag.

EIST can give just as low a wattage, frequency, and core voltage as C1E, except the processor isn't in the halted state. So what does Intel mean?

GeneO · 01 Mar 2012

Live chat with Intel says it is because they are on by default and haven;'t been tested with them off long term. CYA I guess

profdlp · 01 Mar 2012

GeneO said:

...CYA I guess...

I would suspect that you are correct about that.

Of course, a good argument could be made that it has been tested. All you would have to do is look at CPU failure rates for computers under continuous heavy CPU load. They would never have the opportunity to drop down to a lower power state to begin with, making any BIOS C-State settings irrelevant.

I tried to find some statistics, thinking I could compare desktops and workstation CPU failure rates to the CPU failure rates for servers, but had no luck. A general search indicates that when it comes to tracking common causes of computer failure when broken down (no pun intended, but I should have!) by component, CPUs don't even seem to merit a mention.

List of hardware failure rates

PC Hardware Failure Rates By Component and Manufacturer | Geek Montage

As for personal experience, I worked in computer repair for quite a few years and don't ever remember having to replace a CPU which just "broke". Fried by HSF failure, yes. Just worn out from overuse, no.

GeneO · 01 Mar 2012

profdlp said:

GeneO said:

...CYA I guess...

I would suspect that you are correct about that.

Of course, a good argument could be made that it has been tested. All you would have to do is look at CPU failure rates for computers under continuous heavy CPU load. They would never have the opportunity to drop down to a lower power state to begin with, making any BIOS C-State settings irrelevant.

But that is not the same thing I am talking about. Under high load your voltage and temperatures are high. High temperature and voltage will age a processor. I am talking about disabling the low power C1E, C3 and C6 states so the processor only runs in C0 (running state - well maybe if C1E is disabled it drops to C1 with no freq or voltage lowering) with EIST. With EIST it will run at lower multiplier and frequency when not under load, so it is considerably different than a high load CPU running at high volts and freq.

profdlp · 01 Mar 2012

I got you, and I even read a summary of all the C-States and their history so I wouldn't be quite as fuzzy about them before posting. (Not that I'm claiming to suddenly be an expert, either.)

My point was that there is a big difference between ~~African and European Swallows~~ running a CPU at "stock" and basically underclocking it, regardless of the load. If I had been able to find any stats at all it probably would have been best to have three - no, four! - categories:

1) C-Whatever enabled, typical use. (Meaning generally a lot closer to idle than full load.)
2) C-Whatever enabled, heavy use. (Servers, dedicated Folding at Home rig, etc.)
3) C-Whatever disabled, typical use.
4) C-Whatever disabled, heavy use.

I used the example of a server because it took the C-State out of the equation. The fact that several searches using different search terms found no results indicating that CPUs were dropping dead left and right in server rooms under heavy load would lead me to believe that running full speed with the variable load you're apt to see with home use would not likely be a problem.

I am talking about desktop/workstation computers, by the way. Along with extended battery life, the excessive heat in many laptops certainly makes me a fan (the puns are pretty near automatic for me tonight) of leaving it enabled in a laptop.

GeneO · 01 Mar 2012

I don't want to mince words but the server example isn't relevant to my question. Also I am not talking about running it at stock and letting it downclock but running it overclocked and letting it downclock when not under load - with the C states disabled and EIST enabled. The only one that is is your number 3 and only f you add EIST, for without it or the C states, the cores won't downclock.

I am not concerned about laptops, but keeping my desktop relatively cool and at low voltage while idle without affecting the performance and responsiveness when I need it. Using just EIST give me control of that and lets me specify the lowest multiplier. When C states are enabled you no longer have that control.

profdlp · 01 Mar 2012

Perhaps I'm oversimplifying things. (And I am not trying to argue with you - just maybe learn something.) :)

My understanding of C-States is that they are a power-saving measure. With reduced power would come expected greater processor longevity. I think that is the reason for Intel's statement about "Long term reliability..." Since processors aren't croaking under conditions where they can't really take advantage of that (heavy load for long periods of time), the only real benefit I see would be reduced power consumption. Since you're getting that through EIST alone I am not clear about what advantage the C-States may offer.

Am I reading this thing right? Or am I still missing something? :)

GeneO · 03 Mar 2012

Yes your missing my response to myself after chatting with Intel that it is just CYA. I think that was the end of the story

.

I expect a processor using EIST with C-states disabled will be reliable for just as long as with them enabled since the core voltage, temperature and power when not under load are near enough either way,.