C1E, C3 and C6 states, and EIST on Sandy Bridge 2500k

[tetracycloide]

I'm working on getting a good 24/7 OC for my 2500k on my GA-P67A-UD7 board and I'd like to enjoy some of the power savings that these features offer because I like to keep my machine on 24/7. I'm not shooting for the best OC my processer can handle though, I just want something extra for day to day use. I'm at 45x right now (don't have the guts to push it higher yet), 1.42 V CPU under load (set to auto in BIOS), temps never hit 70 on any core even after 17+ hours of Prime 95 small or large LLTs. From what I've read elsewhere these features are often recommended as something to disable when overclocking since they can lead to system instability.

I have two questions:

The risk I think I am taking by enabling these features is system instability at Idle or when switching from Idle to Load or from Load to Idle. I don't think it risks processor degradation or damage to any components at all. Is these assessment correct?

I've been using Prime95 to test my overclocks so far. Is that enough or is there something else I should be doing to specifically look for BSOD or freezes as a direct result of enabling one or more of these features and overclocking at the same time?

 

[bearclaw99]

1.42v seems way too high for 4.5 ghz. Try manually tuning your vcore in BIOS. At least get it under 1.4v

My i7 2600K is running @ 4.5 with 1.304 vcore. Prime 95 seems to be the best test for SB. Intel Burn test doesn't seem to stress it that much in any setting, although that may change with Windows 7 sevice pack 1 which supports AVX

 

[tetracycloide]

I haven't tried to tweak the voltage at all yet but I was going to before I left it alone for good. I've seen much heavier voltage elsewhere so considering the temps are still quite low I wasn't to worried about it at the moment. Are you saying I should get a lower voltage first with the CPU features off and then start playing with turning them on?

 

[bearclaw99]

From what I've seen, it isn't recommended to run it 24/7 over 1.4v (although I am not positive that is the case). As for the C-States, some boards recommend turning them off if you OC, some don't. Mine are all on. All in all, your temps are fine though. Mine hit 75 C on a long P95 run since my cooling could be a bit better, and I think the hyperthreading on the 2600 adds some heat too

But you shouldn't have a problem getting under 1.4v. Auto overclocking tends to raise the voltage to a good deal more than is needed, especially if you don't have a newer BIOS

 

[tetracycloide]

I thought the stock VID for 4.5-4.8GHz was 1.375v-1.450v? It's on the high side of the volt range but on the low side of the clock so yeah, it needs tweaking. I'm aware Auto settings for voltages in the BIOS is basically cruise control and I fully intend to tweak the voltages before leaving the settings on 24/7. That's not what I'm asking about though

 

[Leaps-from-Shadows]

The Sandy Bridge processors are designed to work with those features on, even when overclocked. Turning them off can give you hard drive performance issues, especially when using a SATA III drive. Unless you are at the absolute ragged edge of stability and turning them off keeps you stable, then I would keep the settings on at all times.

Oh, and Intel has confirmed (through their Sandy Bridge processor technical PDFs and online chats) that the maximum is 1.52v for 'safe' Vcore.

bearclaw:
IntelBurnTest actually stresses the CPU more than Prime95. My temps for IBT are higher by 5-10ºC. Of course, I burn it at Maximum stress setting.

 

[RJR]

Yes, you can leave all the Eist, C1E, C-states enabled without a problem on SB.

I've only been playing with my 2600K for a few days (mostly benches) but I recently set turbo multi's to 47x46x45x45x and that only required 1.30v load vcore, so I completely agree with Bearclaw that you really shouldn't need that much voltage.

 

[tetracycloide]

Best answer selected by tetracycloide.

 

[bearclaw99]

Leaps: You do? That's odd, I get 3-4 C higher on a Prime 95 run, and I've run Intel Burn at most levels including Maximum, on Auto-thread and 8 thread

 

[tetracycloide]

In my tests for stability I've found the IntelBurn and Prime95 both generate roughly the same amount of heat, maybe +/- 2 degrees depending on the core.

However, for testing stability with C1E enabled IntelBurn seems to do a much better job because it was designed to run in short bursts. If there's a stability issue being caused by the C1E stepping the best way I've found to find it so far is running 5 run IntelBurn Tests in batches of 5-10. Normally the problem manifests when the message appears that says the system is stable. Almost immediately afterward, when the multiplier is dropping to save power, I'll get a lock up or a BSOD (ironically with a message still on the screen that says 'your system is probably stable).

 

[Leaps-from-Shadows]

That's weird. Like I said, my temps are 5-10ºC higher on each core with IBT. Those tendencies were noted both at stock speed and when overclocked to 4.8GHz. I keep all the EIST, C1E, and CStates enabled. I don't do the short run tests with IBT ... I usually use 50-run Maximum stress settings to do the stress testing while I'm at work. I have 12GB of RAM, so it's testing around 10-11GB of it, and it takes roughly 7.5 hours or so to complete the stress testing.

I've had stability issues when manually overclocking, when the computer would freeze and BSOD while idle (usually when coming down from any sort of load, like tetra said), but not while using the Turbo overclocking presets in the ASRock's UEFI. Those have been rock-solid stable. With my activity, the core speeds are going up and down like mad so they've had plenty of chances to show me instability.

 

[tetracycloide]

I'm not too familiar with the inner works of IntelBurnTest but the impression I get from the interface is that the stress level of the test is directly proportional to the amount of RAM being used. Maybe your temps are significantly higher in IntelBurn because you have 12 gigs of RAM? I've only got 4 gigs.

 

[Leaps-from-Shadows]

I think you've got it right, and I'm glad for it. I'd rather find out that an overclock isn't stable by thrashing it to the max this way than finding out later by having corrupt data or some such.

Thanks for relieving my obsessive brain about this -- it would have bugged me until I came up with an explanation, and that might have taken a while (despite it staring me in the face every time I start IBT up).

 

[tetracycloide]

Just in case anyone is still worried about my voltage I did the tweaking today. Current settings (BIOS):

Multiplier: 45
VCORE: 1.285
Load Line Calibration: Level 2
All C-states and EIST set to Enabled

On Auto the voltage would drop with the C1E stepping when the system was idle to ~1.02 but jump up to 1.42 at load. Now it's 1.308 at idle, 1.320 at load. What I find amazing is that even though the idle voltage is much much higher than before the idle Wattage reading is 4-5 Watts lower. The CPU idles at 6-8W now!

Temps, naturally, improved as well. Peak after 1.5 hours of IntelBurnTest is now 56-63. When I have more time I'll run a full 18+ hour test but these results are phenomenal so far.

 

[bearclaw99]

Just in case anyone is still worried about my voltage I did the tweaking today. Current settings (BIOS):

Multiplier: 45
VCORE: 1.285
Load Line Calibration: Level 2
All C-states and EIST set to Enabled

On Auto the voltage would drop with the C1E stepping when the system was idle to ~1.02 but jump up to 1.42 at load. Now it's 1.308 at idle, 1.320 at load. What I find amazing is that even though the idle voltage is much much higher than before the idle Wattage reading is 4-5 Watts lower. The CPU idles at 6-8W now!

Temps, naturally, improved as well. Peak after 1.5 hours of IntelBurnTest is now 56-63. When I have more time I'll run a full 18+ hour test but these results are phenomenal so far.

There ya go, much better