Delidded my i7-3770K, loaded temperatures drop by 20°C at 4.7GHz

Page 26 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Ed1

Senior member
Jan 8, 2001
453
18
81
I don't know on older cpu but i noticed ivy bridge is very dynamic with temps registered . Evne at stock clocks you get spikes and sometimes in odd times (IMO ) . I noticed during some games I get spike in GUI of game or when not in actual 3d action . And of course you see temps vary according to load and size in something like prime95 .

I use OpenHardwareMonitor which has graph plot and max spike is always higher than avg high temp core runs by like 5c or so (this on stock or just slightly OC with 3570k ).

I guess the faster the polling in the graph it might pick/show it on graph if it was fast enough .

I never noticed such dynamic temps with older cpus like core due , or older cpu's , probably has to do with speedstep and turbo being so dynamic .
 

Ferzerp

Diamond Member
Oct 12, 1999
6,438
107
106
Hi Shmee,

IBT (and likewise LinX) won't know to use only physical cores versus virtual cores, so what you must do is:

1) Set IBT (or LinX) to only use 4 threads from within the setup menu on IBT/LinX.

2) Using task manager, lock the core affinity of IBT/LinX to just one thread per physical core. See this link for a detailed walkthrough.

3) Once IBT/LinX is running the stress test, use task manager and resource monitor to confirm that you are at 50% CPU utilization, if you are not then you need to reassign the core affinity to get around Windows poorly implemented core parking functionality.

Here is a post with proof that you will get higher temperatures.

LinX4threadvs8threads.png

I have to disagree with setting affinity.

The process may not know, but the OS scheduler does and keeps them on the proper logical processors.
 

Yuriman

Diamond Member
Jun 25, 2004
5,530
141
106
IDC, I've been playing around with LLC on my board some more. What I've found is that a setting higher than level 3 (5 being lowest/off, 1 being highest) results in serious voltage overshoot. For example, using +101mv offset with LLC @5 results in a 1c load of 1.304v and a 4c load of 1.248v. With LLC @2, 1c = 1.336v while 4c = 1.352v. I had previously only been trying with LLC at the highest setting, and now it's quite obvious why it would crash with partial loads.

However, even with LLC@3 where I have vdroop, I need a higher voltage to get stable than with LLC off. With LLC@5 I am stable loaded at 1.248v, while with LLC bumped up to 3, I need a loaded voltage of around 1.264v or I get BSODs. The advantage is, on a small load (1 core blend prime) my voltage is 1.272v instead of 1.304v.

You gave your opinion on this before, but I'd like to ask again, do you think it's worth raising my heavy load vcore up from 1.248v to 1.264v in order to bring my light load vcore down from 1.304v to 1.272v?
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
IDC, I've been playing around with LLC on my board some more. What I've found is that a setting higher than level 3 (5 being lowest/off, 1 being highest) results in serious voltage overshoot. For example, using +101mv offset with LLC @5 results in a 1c load of 1.304v and a 4c load of 1.248v. With LLC @2, 1c = 1.336v while 4c = 1.352v. I had previously only been trying with LLC at the highest setting, and now it's quite obvious why it would crash with partial loads.

However, even with LLC@3 where I have vdroop, I need a higher voltage to get stable than with LLC off. With LLC@5 I am stable loaded at 1.248v, while with LLC bumped up to 3, I need a loaded voltage of around 1.264v or I get BSODs. The advantage is, on a small load (1 core blend prime) my voltage is 1.272v instead of 1.304v.

You gave your opinion on this before, but I'd like to ask again, do you think it's worth raising my heavy load vcore up from 1.248v to 1.264v in order to bring my light load vcore down from 1.304v to 1.272v?

I'm hestitant to provide guidance at this point because we are missing one more set of crucial data points - the voltage under real-world use scenarios.

Take the two stable cases (1.248/1.304 vs. 1.264/1.272) and see what kind of voltage you are applying to the CPU when running regular apps that you intend to run with your system. If you plan to F@H then load that and see what vdroop is like, if you plan to watch youtube videos then do that and see what vdroop is like, etc.

Once you know how those two different stable LLC setups impact real-world voltage scenarios then we can better make an informed decision with the tradeoff's.

I really wonder what kind of magic unicorn dust ASUS spread on my MIVE-Z that it just seems to nail the Vdroop, regardless of load (it is like a dynamic vdroop that is continually monitored and tuned as the load varies?). I know the mobo is stupid expensive and if that is part of the reason why it is so expensive then I guess at least I got what I paid for.

I have an asrock extreme-6 (I think that is what it is) sitting in its box still, I should drag that out and see what its LLC is like.
 

C.C.

Member
Aug 21, 2012
28
0
0
My post had nothing to do with turning HT off. I have never turned off HT, not even once, in all my years of testing.
I am a tool lol..I didn't mean to sound offensive, but when I did read the following quote below, my tired brain read this as turn HT off in the BIOS to get the max temps vs leaving HT on..I actually was told to turn it off by a member over @ [H], so we could compare my temps to his 3570K at the same speed de-lidded..I copied and pasted part of my post there to here, hence the mix up:oops:..

For max GFlops and max heat generation you want to set IBT to run with only 4 threads.
I am curious to see your results with the ASROCK Exteme 6..I have the Extreme 4, and from what I can tell they use the exact same power circuity/phases etc...The difference seems to be the extra PCI-E 16X slots IIRC..

I am currently still @ 4.8Ghz, with every voltage on "Auto" except for Vcore which is set @ 1.224V using the fixed option..I opted for the fixed option since I have all the power saving features off, and this box will run @ 100% load for Folding once I get the max o/c and voltages worked out.. While I have overclocked every cpu I've ever owned, going back to the first Pentium days, I have no experience with using voltage offsets and even using LLC honestly..I need to read up more on it and listen to people like IDC who are in the know..

I gotta say I was really hoping for 5Ghz, and after my testing, I personally believe that ANYONE claiming a stable 5Ghz o/c without Phase change cooling is simply full of shit..
 
Last edited:

C.C.

Member
Aug 21, 2012
28
0
0
*Snip*I don't know on older cpu but i noticed ivy bridge is very dynamic with temps registered . Evne at stock clocks you get spikes and sometimes in odd times (IMO ) . I noticed during some games I get spike in GUI of game or when not in actual 3d action . And of course you see temps vary according to load and size in something like prime95 .

I never noticed such dynamic temps with older cpus like core due , or older cpu's , probably has to do with speedstep and turbo being so dynamic . *Snip*


ED, it is funny you mention this, as I was wondering the exact same thing..I noticed it when I stressed my cpu @ 3.7Ghz with all 16GB of ram for 25 passes to just make sure everything was working like it should before overclocking..While the load temps over time were very, very nice, I did notice in the log that I would see one core spike up 10-26C for a split second..I assumed it was a bug or some funky issue, but when I went back and looked @ the logs of my i970 @ 4.2Ghz, I noticed they were consistent throughout the same hours and hours of testing..o_O

I really wonder what kind of magic unicorn dust ASUS spread on my MIVE-Z that it just seems to nail the Vdroop, regardless of load (it is like a dynamic vdroop that is continually monitored and tuned as the load varies?). I know the mobo is stupid expensive and if that is part of the reason why it is so expensive then I guess at least I got what I paid for.

IDC, I myself have grown used to that magic dust from Asus over the last 3 MB's I've owned..I used to always purchase high end boards, but honestly never used the million PCI-E slots etc, so this time I decided to get a cheaper board (ASRock Exteme 4)..Part of this reasoning was due to the excellent review/praise that Anandtech gave it, especially with the lack of voltage ripple under high overclocks that other more pricey boards exhibited..I am starting to wonder if I picked a bad time to cheapen out a bit..
 
Last edited:

Ed1

Senior member
Jan 8, 2001
453
18
81
ED, it is funny you mention this, as I was wondering the exact same thing..I noticed it when I stressed my cpu @ 3.7Ghz with all 16GB of ram for 25 passes to just make sure everything was working like it should before overclocking..While the load temps over time were very, very nice, I did notice in the log that I would see one core spike up 10-26C for a split second..I assumed it was a bug or some funky issue, but when I went back and looked @ the logs of my i970 @ 4.2Ghz, I noticed they were consistent throughout the same hours and hours of testing..o_O



IDC, I myself have grown used to that magic dust from Asus over the last 3 MB's I've owned..I used to always purchase high end boards, but honestly never used the million PCI-E slots etc, so this time I decided to get a cheaper board (ASRock Exteme 4)..Part of this reasoning was due to the excellent review/praise that Anandtech gave it, especially with the lack of voltage ripple under high overclocks that other more pricey boards exhibited..I am starting to wonder if I picked a bad time to cheapen out a bit..
when , I first installed the 3570k system , I first thought something was wrong with HS mounting or grease amounts (even though I mounted HS once and checked coverage and evenness) .
I saw my core3 was like 8c or so hotter than rest, which were closer in temps, maybe delta of 3-4c . I then checked reviews and saw many others had one core higher by good amount too . So it must be either flatness of IHS or gap between core or even maybe where graphic and memory controller is in position of core .

Now i am tempted to remove HS and check flatness of IHS and HS, I only did a visual look at install time (looked good from first glance ) .
 

C.C.

Member
Aug 21, 2012
28
0
0
when , I first installed the 3570k system , I first thought something was wrong with HS mounting or grease amounts (even though I mounted HS once and checked coverage and evenness) .
I saw my core3 was like 8c or so hotter than rest, which were closer in temps, maybe delta of 3-4c . I then checked reviews and saw many others had one core higher by good amount too . So it must be either flatness of IHS or gap between core or even maybe where graphic and memory controller is in position of core .

Now i am tempted to remove HS and check flatness of IHS and HS, I only did a visual look at install time (looked good from first glance ) .

While I wish I could blame it on either the IHS or my waterblock not being perfectly flat, I cannot since I lapped the WB completely flat using up to 2000 grit wetsanding (and checked 100 times with a razor blade!!!, and removed the IHS completely..

All my temps are within a 1-2C range except for the first core, and the reasoning behind that is it is the core that is directly under the "inlet" fitting on my cpu block..I tested this theory with my i970 by rotating the block 180, and then a different core was consistently lower..

When you were referring to a single core randomly spiking, I thought we were talking about the same thing, with a single core being consistent for hours and then randomly spiking a good deal higher (15-20C)..o_O
 

Ed1

Senior member
Jan 8, 2001
453
18
81
While I wish I could blame it on either the IHS or my waterblock not being perfectly flat, I cannot since I lapped the WB completely flat using up to 2000 grit wetsanding (and checked 100 times with a razor blade!!!, and removed the IHS completely..

All my temps are within a 1-2C range except for the first core, and the reasoning behind that is it is the core that is directly under the "inlet" fitting on my cpu block..I tested this theory with my i970 by rotating the block 180, and then a different core was consistently lower..

When you were referring to a single core randomly spiking, I thought we were talking about the same thing, with a single core being consistent for hours and then randomly spiking a good deal higher (15-20C)..o_O

no its not random here, but the one core does spike more than others it seems and runs avg hotter than others .
I am all stock setup 3570k with CM 212evo HS on asus P8z77v-pro , I am OC 40, 40, 39, 38 multipliers per core (basic mild 200mhz over stock turboboost ).
I have it set mild OC cause i don't like adding much heat, this is only few c over stock .
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Yeah you can't avoid having the core temperatures be different from core to neighboring core because the die itself is not symmetric. Some cores are next to very cool parts of the die, which helps keep the core itself cooler, while some cores are surrounded by already warm cores which then makes that core all the hotter.

SandyBridgeTempsandDarkSilicon.jpg
 

chimaxi83

Diamond Member
May 18, 2003
5,649
61
101
Dude, if you're "hitting a wall" at 1.6V, you flirting with death lol. 1.6V on water got me nearly 5.6GHz, for a couple runs. If you want more, you'll need LN2.
 

hokies83

Senior member
Oct 3, 2010
837
2
76
1.6 volts 24/7? Planning on keeping that chip for a couple months?

1.6v is a lil more then IB max safe of 1.55..

Id say it would last afew years like this...

Im testing now if i can drop Vcore.. I know 1.6v will not go for 5.2ghz...

I am hopeing to settle around 1.58ish..

Also my max temps are in the 80c range so im no where near the 95c limit. :cool:

Was wondering if messing with blk can help you aswell i heard it can why im asking in here ;)
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
1.6v is a lil more then IB max safe of 1.55..

I'm curious how you came by this info. To my knowledge Intel has not released a max voltage spec for Sandy Bridge, let alone one for Ivy Bridge.

But if they did I would expect the max spec for SB to be less than 1.55V, and much less than 1.55V for IB.

If official numbers are ought there though I'd definitely like to know them.

Also my max temps are in the 80c range so im no where near the 95c limit.
The limit is 105C.

For IB there doesn't seem to be much in the way of secret tweaks to get extra clockspeed. Basically it comes down to voltage and temperature. Spend a bunch of time optimizing your LLC values so that you can lower your voltage without compromising stability, and do what you can (delidding, etc) to get your temperatures as low as possible (because that helps lower voltage even more).

I've seen people report that changing their CPU PLL voltage helped keep their CPU stable at even lower CPU voltages but when I tried that it didn't make any bit of difference. So that may be a YMMV situation (changing CPU PLL voltage). It may help, it might not help.
 

Ed1

Senior member
Jan 8, 2001
453
18
81
I'm curious how you came by this info. To my knowledge Intel has not released a max voltage spec for Sandy Bridge, let alone one for Ivy Bridge.

But if they did I would expect the max spec for SB to be less than 1.55V, and much less than 1.55V for IB.

If official numbers are ought there though I'd definitely like to know them.


The limit is 105C.

For IB there doesn't seem to be much in the way of secret tweaks to get extra clockspeed. Basically it comes down to voltage and temperature. Spend a bunch of time optimizing your LLC values so that you can lower your voltage without compromising stability, and do what you can (delidding, etc) to get your temperatures as low as possible (because that helps lower voltage even more).

I've seen people report that changing their CPU PLL voltage helped keep their CPU stable at even lower CPU voltages but when I tried that it didn't make any bit of difference. So that may be a YMMV situation (changing CPU PLL voltage). It may help, it might not help.

Someone posted this guide, I don't know if it is accurate but list Intel max voltages .

http://www.overclockers.com/forums/showthread.php?t=705818

Maybe were hokies83 is getting values from .
 

Idontcare

Elite Member
Oct 10, 1999
21,118
58
91
Someone posted this guide, I don't know if it is accurate but list Intel max voltages .

http://www.overclockers.com/forums/showthread.php?t=705818

Maybe were hokies83 is getting values from .

I don't see anything from Intel there.

I see some anonymous internet dude claiming he knows what Intel's recommended voltage ranges are based on the method of cooling :confused:

Is that really all it takes for people to fall hook, line, and sinker?

The American President (1995) said:
Lewis Rothschild: They don't have a choice! Bob Rumson is the only one doing the talking! People want leadership, Mr. President, and in the absence of genuine leadership, they'll listen to anyone who steps up to the microphone. They want leadership. They're so thirsty for it they'll crawl through the desert toward a mirage, and when they discover there's no water, they'll drink the sand.

President Andrew Shepherd: Lewis, we've had presidents who were beloved, who couldn't find a coherent sentence with two hands and a flashlight. People don't drink the sand because they're thirsty. They drink the sand because they don't know the difference.

Anonymous internet dude != Intel official spec

I fear for people who are being led to believe their IB chips can take 1.5V on air and live more than 6 months.

Bloomfield (45nm) was officially spec'ed by Intel as having a max Vcc of 1.55V.

Gulftown (32nm) was officially spec'ed by Intel as having a max Vcc of 1.40V.

Intel never specified a max voltage for 32nm Sandy Bridge, but since it used the exact same transistor and metal wiring you would not expect the max safe voltage for the internal spec to be any higher than that of Gulftown.

To my knowledge Intel has not gone on record to specify a max voltage for any of their 22nm products, which is why I am leery of anyone claiming to know the spec limit.

But given what we know to be true of device physics and node shrinks, we would not expect the max safe voltage for the internal spec at 22nm to be higher than the max value considered to be safe for 32nm. Even though the xtors have radically changed, the M1 wiring has not.

In fact the M1 wiring has gotten smaller and denser with 22nm versus 32nm, which would necessitate a decrease in the maximum allowed voltage for the IC to maintain its intended lifespan.

So I would be surprised if the max voltage for IB from an Intel internal spec standpoint is higher than 1.3V, personally. But I see no way to formulate logical argumentation to the effect that 22nm max voltage would be higher than that considered acceptable for the max voltage of 45nm. That is just wishful thinking IMO.
 

Ed1

Senior member
Jan 8, 2001
453
18
81
I don't see anything from Intel there.

I see some anonymous internet dude claiming he knows what Intel's recommended voltage ranges are based on the method of cooling :confused:

Is that really all it takes for people to fall hook, line, and sinker?



Anonymous internet dude != Intel official spec

I fear for people who are being led to believe their IB chips can take 1.5V on air and live more than 6 months.

Bloomfield (45nm) was officially spec'ed by Intel as having a max Vcc of 1.55V.

Gulftown (32nm) was officially spec'ed by Intel as having a max Vcc of 1.40V.

Intel never specified a max voltage for 32nm Sandy Bridge, but since it used the exact same transistor and metal wiring you would not expect the max safe voltage for the internal spec to be any higher than that of Gulftown.

To my knowledge Intel has not gone on record to specify a max voltage for any of their 22nm products, which is why I am leery of anyone claiming to know the spec limit.

But given what we know to be true of device physics and node shrinks, we would not expect the max safe voltage for the internal spec at 22nm to be higher than the max value considered to be safe for 32nm. Even though the xtors have radically changed, the M1 wiring has not.

In fact the M1 wiring has gotten smaller and denser with 22nm versus 32nm, which would necessitate a decrease in the maximum allowed voltage for the IC to maintain its intended lifespan.

So I would be surprised if the max voltage for IB from an Intel internal spec standpoint is higher than 1.3V, personally. But I see no way to formulate logical argumentation to the effect that 22nm max voltage would be higher than that considered acceptable for the max voltage of 45nm. That is just wishful thinking IMO.

I only posted that link cause it seemed to match vcore max hokies83 was saying .
I agree that seems way high . I guy from Asus said to keep below 1.4v and shoot for 1.3v area on air/water . That is from vids that they posted on newegg and you-tube for marketing there MB line .
 

Ferzerp

Diamond Member
Oct 12, 1999
6,438
107
106
Yeah you can't avoid having the core temperatures be different from core to neighboring core because the die itself is not symmetric. Some cores are next to very cool parts of the die, which helps keep the core itself cooler, while some cores are surrounded by already warm cores which then makes that core all the hotter.

SandyBridgeTempsandDarkSilicon.jpg

My last two cores are actually the same temp (and through all the remounting I did over time, were always so) I'm not sure why. The first two are quite a bit lower (more than 10C for the first)

Just checked last night, and peaks still 68C with IBT.
 

ehume

Golden Member
Nov 6, 2009
1,511
73
91
Bloomfield (45nm) was officially spec'ed by Intel as having a max Vcc of 1.55V.

In support of what you're saying, IIRC the Lynnfield max Voltage spec was initially 1.55v, then later revised to 1.4v. It's certainly 1.4v now.
 

Ferzerp

Diamond Member
Oct 12, 1999
6,438
107
106
I don't see anything from Intel there.

I see some anonymous internet dude claiming he knows what Intel's recommended voltage ranges are based on the method of cooling :confused:

Is that really all it takes for people to fall hook, line, and sinker?



Anonymous internet dude != Intel official spec

I fear for people who are being led to believe their IB chips can take 1.5V on air and live more than 6 months.

Bloomfield (45nm) was officially spec'ed by Intel as having a max Vcc of 1.55V.

Gulftown (32nm) was officially spec'ed by Intel as having a max Vcc of 1.40V.

Intel never specified a max voltage for 32nm Sandy Bridge, but since it used the exact same transistor and metal wiring you would not expect the max safe voltage for the internal spec to be any higher than that of Gulftown.

To my knowledge Intel has not gone on record to specify a max voltage for any of their 22nm products, which is why I am leery of anyone claiming to know the spec limit.

But given what we know to be true of device physics and node shrinks, we would not expect the max safe voltage for the internal spec at 22nm to be higher than the max value considered to be safe for 32nm. Even though the xtors have radically changed, the M1 wiring has not.

In fact the M1 wiring has gotten smaller and denser with 22nm versus 32nm, which would necessitate a decrease in the maximum allowed voltage for the IC to maintain its intended lifespan.

So I would be surprised if the max voltage for IB from an Intel internal spec standpoint is higher than 1.3V, personally. But I see no way to formulate logical argumentation to the effect that 22nm max voltage would be higher than that considered acceptable for the max voltage of 45nm. That is just wishful thinking IMO.


It is likely that someone took the range of values for VID and assumed that those were ok values, which is kind of silly. I do think 1.55V is the highest value for that register though. This is no way suggests that the voltage is ok.

Well, looks like that could only go up to 1.52V per http://www.intel.com/content/dam/ww...eets/3rd-gen-core-desktop-vol-1-datasheet.pdf
 
Last edited:

BonzaiDuck

Lifer
Jun 30, 2004
15,699
1,448
126
It is likely that someone took the range of values for VID and assumed that those were ok values, which is kind of silly. I do think 1.55V is the highest value for that register though. This is no way suggests that the voltage is ok.

We've seen here and there, some eager, young or otherwise less disciplined folk who haven't even read and studied their motherboard manual. You'd even wonder if some would ever bother scanning through Intel's own web-pages, spec-sheets, and other available information.

Frankly, I don't even trust to my own conclusions about some things without soliciting "second opinions" with a well-stated question, and then critiquing those opinions. Heck. That's why I stay plugged in to these forums.