i7-2600k oc voltages

[Sunburn74]

Hey so I want to run something by you guys.

I have a i7-2600k a ga-p67x-ud and 16gb corsair vengeance ram. Right now to run the chip at 4.4ghz, I need about 1.26-1.27 v (0.4 v below stock voltage). To run the chip at 4.6 i need about 1.33v (0.15 v above stock voltage). WIth that my load goes up about 4-5 deg celcius

Do you guys think its worth the voltage and heat to go up to 4.6 ghz? Or am I doing something wrong?

With my oc all power features are on and so is hyperthreading. The only thing I really need to adjust is voltage vcore specifically using the dvid method. Pretty much everything else is on auto.

 

[choliscott]

I don't own this particular cpu. You mentioned that you need about 1.26 to 1.27 to get to 4.4ghz. Do you know how much you need to get to 4.5ghz?

While I believe that 1.33v is still fine, you have to decide if that much of an increase for 200mhz is worth it. Another question, have you tried going higher then 4.6ghz? You might have one of "golden chips"

Hey so I want to run something by you guys.

I have a i7-2600k a ga-p67x-ud and 16gb corsair vengeance ram. Right now to run the chip at 4.4ghz, I need about 1.26-1.27 v (0.4 v below stock voltage). To run the chip at 4.6 i need about 1.33v (0.15 v above stock voltage). WIth that my load goes up about 4-5 deg celcius

Do you guys think its worth the voltage and heat to go up to 4.6 ghz? Or am I doing something wrong?

With my oc all power features are on and so is hyperthreading. The only thing I really need to adjust is voltage vcore specifically using the dvid method. Pretty much everything else is on auto.

 

[MrTransistorm]

WIth that my load goes up about 4-5 deg celcius

Relative to what? We need a frame of reference here.

As for using power-saving features, turbo, HT, and DVID, by all means use them if you can. You shouldn't have to disable any of them unless you are going for a very high overclock.

 

[BonzaiDuck]

WIth that my load goes up about 4-5 deg celcius

Do you guys think its worth the voltage and heat to go up to 4.6 ghz? Or am I doing something wrong?

DITTO for MrTransistorm's "Relative to what . . . ?"

Much has been written here at the forums recently on this topic. The TM1 thermal-spec for the i7-2600K is well-specified; the voltage "safe-range" is not so clear, but we have a consensus here at the forums that you should be able to run at least an IDLE voltage for full "turbo" speed of about 1.35V, and some of the gurus say that 1.37V is still "safe." This is based on the die-size of earlier socket-1366 I7's.

Second, you'd think that the LOAD voltage -- which is drooped or lower than IDLE and is the voltage when the processor is under load stress -- might even be more relevant, but I use my idle "turbo-speed" voltage as a measuring stick. Because . . . . I'm a cautious devil . . .

Third, how you measure temperature is also important to your questions, because most of the shareware test-software like RealTemp, CoreTemp measure only core [TJunction] temperatures, while AIDA64 [the old Everest Ultimate] and HWMonitor report the same core temperatures and a "package" temperature. It may be that the package temperature is the specification reference for Intel's TCASE, but but these latter programs seem to have it wrong, because TCASE should always be several degrees below the cores, while the programs can show it equal or higher.

If you're not measuring and monitoring TCASE, or you can't trust the program that reports it, assume it be 5 to 7C degrees below the average core value. Some might say 10C. And some more accurate monitoring software (like ASUS Monitor for ASUS motherboards) might show it to be 10C below the core average.

Are you "turbo-overclocking," or have you set the VCORE to a fixed value and disabled the EIST, C1E, etc.? You don't need to do the latter anymore. If you leave VCORE on "Auto" and adjust "Offset" and "Extra voltage for turbo-mode," you can just as precisely control the idle-to-load voltage-range for the turbo-speed you want. In that case, you would barely see the idle "turbo" voltage; the software monitor may pick it up, but it will only endure for a few seconds at the beginning of processor loading and afterwards for a split second -- just before the processor speed returns to the EIST mode idle speed.

 

[Sunburn74]

Relative to my temp at 4.4

In other words, to go up from 4.4 to 4.6, the increase in voltage results in a temp increase of 4-5 degrees celsius. But I'm not measuring a core temp here. Just what the board reports.

So basically you guys feel like 1.33v is appropriate for a 4.6ghz overclock?

 

[BonzaiDuck]

Relative to my temp at 4.4

In other words, to go up from 4.4 to 4.6, the increase in voltage results in a temp increase of 4-5 degrees celsius. But I'm not measuring a core temp here. Just what the board reports.

So basically you guys feel like 1.33v is appropriate for a 4.6ghz overclock?

Yes. I've had a few exchanges with other here; we wrote some extensive threads on this topic; they were experienced OC'ers, and I'm no virgin to it. In other words, they and I can be trusted to have found minimum rock-stable voltages.

I'm running at 4.6 Ghz; their voltages for 4.6 are near identical to mine. My "drooped" VCORE under full load is either 1.31V or 1.33V, depending on the chosen stress-test. Frankly, I think I can get it lower by just a sliver on the "Offset" setting, since my current voltage settings result in 0.005V lower than the settings I used for 4.62 and/or 4.64 with a slight bCLK overclock to 103 Mhz.

I'm a stickler for "rock-stable," and require at least 5 hours under PRIME95 Large FFT test and 30 iterations of IBT on a "High" to "Maximum" stress setting. If I wasn't using IBT, I'd be inclined to run PRIME95 for 10 hours. Maybe longer . . .

It was never clear to me in this thread whether you were talking about an "idle" voltage at the overclocked speed, or the drooped loaded voltage. If you register in the 1.33V under stress load, you would expect to see an idle voltage at about 1.36V. Although . . . that might depend or vary by individual processor. If you're speaking of your idle voltage with the speed fixed at 4.6, I don't see how you could do it, or -- you got a hellofa good processor.

On the temperature angle, of course you can expect a rise in temperature for increasing the voltage -- whether it's a dynamic VCORE or a fixed setting. But the TM1 throttling spec for that processor is 72.6C for the "TCASE" temperature. As I may have said, this TCASE value should be lower by several degrees from what you see in CoreTemp or RealTemp as the "TJunction" core temperature values.

With my NH-D14 cooler, my TCASE shows to be in the low 60's C, and my core values range between about 65 and 72C. That is, I have one core with a temperature (72C) that is too far from the cluster-average of the other three (which are all close together). So I surmise that I have core sensors of varying accuracy.


IN any case, I could probably bump up to 4.7+Ghz to a point where my TCASE temperature peaks at -- say -- 71C.

 

[grkM3]

Are you seeing 1.33 loaded or 1.33 idle? I think stock vid will call for up to 1.37 volts on the 2600k with 8 threads running full load.

you are 100% safe at anything under 1.37 as long as your temps are under 80c full linx avx load.

you have a good chip if it can do 4.6 full load with 1.33 in bios,that means its like 1.315 under full load.

 

[Idontcare]

Are you seeing 1.33 loaded or 1.33 idle? I think stock vid will call for up to 1.37 volts on the 2600k with 8 threads running full load.

you are 100% safe at anything under 1.37 as long as your temps are under 80c full linx avx load.

you have a good chip if it can do 4.6 full load with 1.33 in bios,that means its like 1.315 under full load.

VID will go all the way up to 1.5V as the CPU's PCU automatically monitors clockspeed, thermals, and load.

Monitoring the VID for my 2600K, I have watched it firsthand go all the way up to 1.3991 Volts (the VID does) at 4.5GHz. I set the offset though so the CPU only gets 1.3336 volts at that clock.

I have not attempted to clock my 2600K above 4.5GHz yet, but when I do I am fully expecting to see the CPU requesting the Vcc be set to something above 1.4V per the VID given how it has scaled so far.

 

[MrTransistorm]

Relative to my temp at 4.4

Yes, but what are your core temps at 4.4 GHz? If they were around say 60C under load, then another 5C is no problem. If they were above 80C at 4.4 GHz, then 4.6 GHz would be pushing it.

 

[tizm]

Relative to my temp at 4.4

In other words, to go up from 4.4 to 4.6, the increase in voltage results in a temp increase of 4-5 degrees celsius. But I'm not measuring a core temp here. Just what the board reports.

So basically you guys feel like 1.33v is appropriate for a 4.6ghz overclock?

I'd say anywhere around 1.35-1.37 vcore is ok, as long as the temps don't reach 80 or mid 70's ideally.

Personally I use 4.5ghz @ 1.34 volts (LLC / vdroop is minimal so highest it will go is like 1.344 and will underclock to 1.33 ish at times) and temps ranging from 65-70. I could push it more but it takes way too much increase in voltage to get 4.7ghz (1.4v and around 70-78) so I opted to stick to 4.5.

If I had your chip, I'd push it a bit further if the temps were good enough.

 

[aigomorla]

thats because VID is linked to multiplier now... lol..

Its been linked since the Yorkfield E0 days.

Dont u guys remember me telling all the cheaters with low vids to RAISE their multi..

oops.. wait they couldnt do that.. because only a handful of us on the forum had real EE cpu's which costed an arm and a leg and allowed u to do that b4 the 2600K!!!

 

[Sunburn74]

.
It was never clear to me in this thread whether you were talking about an "idle" voltage at the overclocked speed, or the drooped loaded voltage. If you register in the 1.33V under stress load, you would expect to see an idle voltage at about 1.36V. Although . . . that might depend or vary by individual processor. If you're speaking of your idle voltage with the speed fixed at 4.6, I don't see how you could do it, or -- you got a hellofa good processor.

So the way I'm overclocking is with the dvid method. This lets voltages be variable according to load on gigabyte boards. Basically I select normal vcore and choose the offset. What that will tell the board is my maximum allowed voltage, whilst keeping the power saving and voltage variability present. To put it more clearly, if you have your board on all auto at completely stock settings, you'll notice that at idle or low loads your vcore will be like 0.8v but at full load you'll be like 1.26v. However when you punch in a vcore of say 1.3, it stays 1.3 (roughly) regardless of load. Thats because you've put in a static voltage and turned off the stock voltage variability. To overclock whilst keeping that voltage variability, you have to overclock using the dvid method. On gigabyte boards for voltage pick normal (you actually select normal, and not auto or a manual entered value) and for dvid choose the increases or decreases that is equivalent to the static value you want. If stock is 1.3v, and I want to run my cpu at 1.25, I choose an offset of (-) 0.05v. If stock is 1.3, and I want to run my chip at 1.4, I choose an offset of +0.1V. And so on. However with both, if my chip is idle and running at those very low speeds the sandybridge processors enter to save power, the voltage will also drop dramatically as the chip was designed to do; when you enter a manual value into the voltage setting that voltage drop for power savings is what you are turning off.

My cpu idles at like 22-24 degrees because at idle or low loads my vcore is like 0.89v and I have EIST, cie, s3/s6 on so my speed is like 1500mhz or less. However when running prime95 or linx, my vcore is roughly 1.26 (which is still 0.4v below the stock value) and my speed ramps up to 4.4ghz. To go to 4.6ghz max, I have to increase my dvid from -0.4 below stock to 0.15v above stock (stock is approx 1.36v). What I'm saying is that after a prime95 run, my temps are reported to be 55-58 degrees celcius at 4.4ghz and 1.26. However to go to 4.6ghz, my temps go up to roughly 62-64 degrees celcius and my vcore is reported to be 1.33. I don't care about the idle because idle is always ridiculous low due to power savings regardless of how high my offset is. its the load temps and load voltages that matter to me. I guess though I need to check what my core temps are. All I'm looking at is what the board is reporting. Still you guys feel like 1.33 isn't that bad so maybe I'll see if I can't get a little but more out of the chip.

Hope that clears it up. Dvid method basically lets you save power at low loads, keep your cpu cooler at low loads, but may sacrifice some overclockability because you do lose some stability as you're getting close to your CPU max speed (after all you're jumping from like .8v to 1.33v.

 

[Idontcare]

So the way I'm overclocking is with the dvid method. This lets voltages be variable according to load on gigabyte boards. Basically I select normal vcore and choose the offset. What that will tell the board is my maximum allowed voltage, whilst keeping the power saving and voltage variability present. To put it more clearly, if you have your board on all auto at completely stock settings, you'll notice that at idle or low loads your vcore will be like 0.8v but at full load you'll be like 1.26v. However when you punch in a vcore of say 1.3, it stays 1.3 (roughly) regardless of load. Thats because you've put in a static voltage and turned off the stock voltage variability. To overclock whilst keeping that voltage variability, you have to overclock using the dvid method. On gigabyte boards for voltage pick normal (you actually select normal, and not auto or a manual entered value) and for dvid choose the increases or decreases that is equivalent to the static value you want. If stock is 1.3v, and I want to run my cpu at 1.25, I choose an offset of (-) 0.05v. If stock is 1.3, and I want to run my chip at 1.4, I choose an offset of +0.1V. And so on. However with both, if my chip is idle and running at those very low speeds the sandybridge processors enter to save power, the voltage will also drop dramatically as the chip was designed to do; when you enter a manual value into the voltage setting that voltage drop for power savings is what you are turning off.

My cpu idles at like 22-24 degrees because at idle or low loads my vcore is like 0.89v and I have EIST, cie, s3/s6 on so my speed is like 1500mhz or less. However when running prime95 or linx, my vcore is roughly 1.26 (which is still 0.4v below the stock value) and my speed ramps up to 4.4ghz. To go to 4.6ghz max, I have to increase my dvid from -0.4 below stock to 0.15v above stock (stock is approx 1.36v). What I'm saying is that after a prime95 run, my temps are reported to be 55-58 degrees celcius at 4.4ghz and 1.26. However to go to 4.6ghz, my temps go up to roughly 62-64 degrees celcius and my vcore is reported to be 1.33. I don't care about the idle because idle is always ridiculous low due to power savings regardless of how high my offset is. its the load temps and load voltages that matter to me. I guess though I need to check what my core temps are. All I'm looking at is what the board is reporting. Still you guys feel like 1.33 isn't that bad so maybe I'll see if I can't get a little but more out of the chip.

Hope that clears it up. Dvid method basically lets you save power at low loads, keep your cpu cooler at low loads, but may sacrifice some overclockability because you do lose some stability as you're getting close to your CPU max speed (after all you're jumping from like .8v to 1.33v.

ASUS mobo's have the same feature, its not called "dvid" though, it's just refered to as "offset".

At stock clocks, my 2600K can function on so little Vcc while fully loaded in LinX that the offset can become quite large.

So large in fact that the system becomes unstable during idle because the large negative offset results in the idle voltage being around 0.7V.

So for my system, I either use excess volts at load or excess volts at idle, I can't have the best of both worlds.

What we need the mobo makers to do is code the bios to let us specify our own fixed Vcc versus multiplier table.

16x = 0.800V
34x = 1.110V
38x = 1.120V
45x = 1.334V

And so on. Kind of like a user-defined fan profile or shmoo plot.

 

[BonzaiDuck]

ASUS mobo's have the same feature, its not called "dvid" though, it's just refered to as "offset".

At stock clocks, my 2600K can function on so little Vcc while fully loaded in LinX that the offset can become quite large.

So large in fact that the system becomes unstable during idle because the large negative offset results in the idle voltage being around 0.7V.

So for my system, I either use excess volts at load or excess volts at idle, I can't have the best of both worlds.

What we need the mobo makers to do is code the bios to let us specify our own fixed Vcc versus multiplier table.

16x = 0.800V
34x = 1.110V
38x = 1.120V
45x = 1.334V

And so on. Kind of like a user-defined fan profile or shmoo plot.

Sure. I just wonder if they don't use some purely linear relationship inputting VID or Vcc to scale the voltage with any "auto" setting. Sunburn clarified what he's doing -- no different than the rest of us.

 

[Sunburn74]

I also want to ask you guys.

If running prime95 to stress test the my PC, if I have 16gb of ram in my box, is prime blend ok? Or should I specify prime95 to use higher amounts of ram?

 

[mrjoltcola]

Another question, have you tried going higher then 4.6ghz? You might have one of "golden chips"

Just curious, are you saying that some chips are stable at certain higher frequencies than lower frequencies? I'd never heard of this.

I couldn't stabilize my 2600K at 4.7 so I didn't go higher.

 

[Idontcare]

I also want to ask you guys.

If running prime95 to stress test the my PC, if I have 16gb of ram in my box, is prime blend ok? Or should I specify prime95 to use higher amounts of ram?

For Prime95, use small FFT if you are trying to test the stability of the core's themselves (core+L1$+L2$), use large FFT if you are trying to test the stability of the IMC, L3$, etc.

Blend is actually the least ideal selection as it alternates between testing small and large FFTs, the result of which means you actually don't stress-test either set of components for any prolongued period of time.

Ideally you would stress test on Large FFT, confirm the memory loop is good, for say 12 hrs, and then you stress test with small FFT to confirm the core logic is stable.

12 small + 12 large != 24 blend. Blend will never push your core temps up and keep them up for the sustained periods of time necessary to actually "stess test" the cores, likewise for large FFT's.

Even more insidious, with Blend you have no idea which portion of your hardware was the weakest link should your rig fail to pass even a blend stability test. With small and large FFT tests you know exactly what you need to go after, be it more cooling or more volts or lower clocks, to stabilize the rig.

PS - if you are stress testing a CPU with hyperthreading then you also need to manually set the thread count to be equal to your CPU's core count. Don't do 8 threads of small FFT on a 2600K, your temps will actually be lower than if you set it to run 4 threads, and that means you won't actually be stress-testing your 2600K to the limits you want to be stressing it at for stability purposes.

Just curious, are you saying that some chips are stable at certain higher frequencies than lower frequencies? I'd never heard of this.

I couldn't stabilize my 2600K at 4.7 so I didn't go higher.

No, it doesn't work that way. In the old days with "FSB holes" it would seem like this was the case, but that was because the weakest link was the NB chipset and not the CPU.

With a 2600K the CPU will be the weakest link assuming you are OC'ing by multiplier. There are no "islands of stability" like transuranium elements on the periodic table. Device physics don't work like that.

 

[mrjoltcola]

Ah, very good, I thought so. Thanks! (Though I was about to crank mine to 5.0 and try it anyway)

 

[Sunburn74]

Just curious, are you saying that some chips are stable at certain higher frequencies than lower frequencies? I'd never heard of this.

I couldn't stabilize my 2600K at 4.7 so I didn't go higher.

Yeah asus looked into it. Google it. Their max allowable voltage was 1.4v i believe. Like 80-90 percent of chips can hit 4.4 needing just stock voltage. 50 percent hit like 4.5 with variable voltage increases. I want to say about 20 percent hit 4.7 . And definitely like 5-10 percent can go to a solid 5.0

There are people out there running rigs with 5.0ghz processors and under 1.4v believe it or not.

@idontcare
You miss my point.

Lets say you have 4gb of ram in your box and you run prime blend. Prime will take up about 3gb and will truly test for the most part the majority of your ram.

Now lets say you have 16gb of ram in your box and you run prime blend. Prime will still only take up about 3gb which is a mere fraction of what you actually have in your box. Is prime blend still testing all your ram?

Or should I manually force prime to use say 12gb or 13gb of ram?

 

[jvroig]

You shouldn't be using Prime Blend to test your RAM.

Prime95 is only good for Small FFT (cpu, more volts or lower clocks) and Large FFT (IMC). Blend gives you no clue what's wrong, and is less "stressful" than either.

If you wanted to test your RAM, use Memtest86+ or MemtestHCI.

EDIT: But to answer your question, if you want more RAM coverage, then yes you should certainly force it to use more. But the point remains - you will be far better off not using Blend and using a real memory tester, so all your tests are as isolated as it can get from other factors.

 

[Sunburn74]

Well the point of prime blend is to test both the cpu and the ram, a mark of total system stability.

 

[jvroig]

That's just it though - when something happens in blend, you aren't sure if it's CPU related or RAM or IMC.

That, and it is less stressful than small fft and large fft.

That's the whole point of IDC's post and my post as well.

If you use small fft, and it reboots or produces an error, you know you either have to lower your clocks, or increase voltage (whatever floats your boat).

If you use large fft, and it produces an error, you know it's the memory subsystem. In an AMD setup (not sure about the Intel counterpart) this would be time to increase your CPU-NB voltage, or lower your CPU-NB/IMC clocks.

Blend... not as stressful as either, and gives you no clue what's wrong (could be CPU or RAM, who knows). It's a two-punch combo that you don't have to live with.

In more concrete terms, you can Blend for 12 hours, and it may pass, but then not pass both small fft and large fft for twelve hours each, simply because the other two tests are more stressful to their respective targets.

 

[Idontcare]

Well the point of prime blend is to test both the cpu and the ram, a mark of total system stability.

That's not what prime95 does though, neither large nor small FFT (of which blend is an admixture of the two) are intended to test your ram.

Only in the extreme cases of really really unstable ram would a large FFT result in an error. But those sorts of memory errors ought to have been fleshed out earlier on with a memtest+ run.

Prime95 is both a stability test as well as a stability diagnostic tool.

The stability test portion is compromised when you run Blend instead of small or large because Blend does not sustain the stress placed on the core logic or the IMC/L3$ owing to its very nature of alternating between small and large FFT's.

Running blend is tantamount to pitching underhand to your rig while stress-testing and then being excited that it hit it out of the park.

Also, because of this, folks are surprised down the road to find their supposedly blend-stable OC is crashing in games and apps, then when they get told to go run prime95 small or large explicitly their rig fails to pass muster.

It's just the nature of the program. Blend was never intended to be as stringent of a stability test as running small and large FFT's explicitly and separately.

edit: lol, I had this window open from last night and had not realized jvroig's response was already added before I drafted my response above...now I see I could have simply did a "+1" to jvroig's post

 

[Sunburn74]

So then what is the point/usefulness of prime blend then?

And how would you recommend prime actually be used to stress test?

 

[Idontcare]

So then what is the point/usefulness of prime blend then?

There is none, other than to say that if your rig won't even pass blend then its got serious stability issues.

You may as well ask me what the point/usefulness of SuperPi is in terms of stressing suicide OC runs.

And how would you recommend prime actually be used to stress test?

If I were limited to stress testing my 2600K using only Prime95 then I would test w/4-threads w/small FFT 8-12 hrs followed by testing w/4-threads w/large FFT for 8-12hrs.

However, I wouldn't recommend using Prime95 now that LinX is available.

For stress testing I use HCI for the ram, Prime95 large FFT for the IMC/L3$, and LinX for the CPU core logic.