4770k delid, mixed results

[BonzaiDuck]

I have not messed with the bclk. I thought that was taboo since Sandy Bridge.

It was taboo WITH Sandy Bridge -- posi-lutely abso-tively. Or -- not so much taboo, but discouraged. We all tried it. Some still do it -- modestly. Take for instance my latest overclock increase and benching to 4.7Ghz. That's a multiplier of 47. At this point, I could get almost 4.75 Ghz (on its way to 4.8) just by upping the bCLK from 100 to 101.

But let's admit that my experience is "great" but not "up-to-date." I couldn't just jump on the SB-E, IB or IB-E bandwagon, because I have practical considerations -- I live within a budget -- and my 2600K was the best bang-for-buck I ever had. "Back when . . " -- I merely read about the bCLK straps, only understanding that there were implications, but not fully tuned-in to what those implications were.

Here's the link to the G.SKILL Forum article on SB-E and X79, as pertains to memory, the "straps" and how or why OC'ing could be "easier" in the 4.4 to 4.9Ghz range:

http://www.gskill.us/forum/showthread.php?t=10512

Now -- before you jump all over this, and all over me, I didn't study it like I did to ace 3rd-semester Organic Chemistry when I was a college sophomore. I just found it, I'm old -- and it was late. So as someone mentioned earlier, reexamine the platform-specific articles on your IB or Haswell choices.

But something just tells me -- even if you don't fiddle with the new bCLK for SB-E, IB/-E and Haswell, it would have "something to do" with stability in overclocking, and possible "walls" which are not necessarily insurmountable.

One more point. I find myself taking "Maximum PC Magazine" with a few grains of salt: they publish for a certain type of target audience, which I may merely overlap. Their December 2013 article in an issue that looks toward 2014 is pushing the IB-E cores for socket-2011 -- despite the dead-end for the socket. Max PC is skeptical that the TIM problem for socket-115x IB and Haswell cores doesn't make a lot of difference, but I thought that IDontCare's lengthy tests and results do show something there.

Now -- tell me I'm an old fart spewing hot air, full of bool-s***, don' know what I'm talkin' about. Partly right there: I haven't kept up since I first OC'd my 2600K except for reading the de-lidding threads.

 

[cytg111]

Pretty much inline with other delid reports, 200-300Mhz gain at most and a 10-20 degree drop in temp. Personally it aint worth the effort, but hey, I dont crave uber-uber 97% will do just fine for me.

 

[thilanliyan]

Quick question, why is it that when I run a stress testing program, the voltage fed to the CPU is way higher than what I set in the BIOS?

I have it at an offset to reach 1.25v, but under P95 for example, it goes to 1.35v. Only way to avoid is to set it manually (no offset) to 1.25v. I was under the impression that offset mode was best to do, rather than setting the voltage manually?

 

[BonzaiDuck]

Quick question, why is it that when I run a stress testing program, the voltage fed to the CPU is way higher than what I set in the BIOS?

I have it at an offset to reach 1.25v, but under P95 for example, it goes to 1.35v. Only way to avoid is to set it manually (no offset) to 1.25v. I was under the impression that offset mode was best to do, rather than setting the voltage manually?

What setting do you have for Load-Line-Calibration or LLC?

Again -- Ah been poh dis yeah -- my rental propitty been vacant. I'm still nursing this Sandy Bridge and happier-than-a-pig-in-s***. So I can only speak for this "Ol' Sandy Bridge" 2600K.

I had been very cautious with LLC. At the beginning, I gave it a boost up to "Regular" or 25%. Consulting with IDC here (and "back in the day") -- it was clear that bumping up to the "High" setting would leave a vDroop of only 20mV, but it was "ideal." "Ideal" for SB overclocking.

the other day, I bumped it up to "Ultra," -- a step below "Extreme," started a stress-test and stopped it in a hurry, because the voltage at load (I think) was pegging around 1.41V.

So . . . . what's your LLC setting? Does this help? Or am I hopelessly stuck in horse and buggy days?

 

[thilanliyan]

^That COULD be it. I have it set at Level 4 (highest is Level 7 I think). But does that really up the voltage by up to 0.1v?

Having said that though, I left that setting alone, changed voltage to manual, and it stays to what I set it at during P95.

What is the advantage of using offset mode anyway? The power saving features work fine using manual voltage from what I can see.

Oh and...PROGRESS!! I'm at 4.1GHz now at 1.25v.

 

[rtsurfer]

Quick question, why is it that when I run a stress testing program, the voltage fed to the CPU is way higher than what I set in the BIOS?

I have it at an offset to reach 1.25v, but under P95 for example, it goes to 1.35v. Only way to avoid is to set it manually (no offset) to 1.25v. I was under the impression that offset mode was best to do, rather than setting the voltage manually?

Because when you are stress testing you are supposed to use manual.

Prime 95/ Aida 64/ Lin X cause unreal strain on the CPU.
If you some real world app like playing BF4 for a few hours.
Your voltage wont go up no matter what.

When your stress testing a new higher OC you use manual.
Once you know the OC is stable,you switch to Adaptive.

Also LLC will effect your load voltage no matter you use manual or adaptive or offset.

Please watch this Video on how to OC Haswell properly

http://www.youtube.com/watch?v=0CHs5_TdpXE

 

[BonzaiDuck]

Because when you are stress testing you are supposed to use manual.

Prime 95/ Aida 64/ Lin X cause unreal strain on the CPU.
If you some real world app like playing BF4 for a few hours.
Your voltage wont go up no matter what.

When your stress testing a new higher OC you use manual.
Once you know the OC is stable,you switch to Adaptive.

Also LLC will effect your load voltage no matter you use manual or adaptive.

Please watch this Video on how to OC Haswell properly

http://www.youtube.com/watch?v=0CHs5_TdpXE

If there's not a newer white-paper on the topic, Anandtech published a paper sometime . . . December 19, 2007 . . . on "overclocking the QX9650 -- the rules have changed." Here's the link:

http://www.anandtech.com/show/2404

This is old stuff! The flagship processor looked toward the Yorkfield line -- heck! -- it was a Yorkfield. But it includes material explaining LLC, offset and other factors. The article's main concern: voltage change exhibits a harmonic process. To use a puerile expression that seems to describe it in idiotic terms, "it goes 'boi-oi-oi-ing!" There are unregistered or unmeasured spikes that occur when processor load ceases, and before voltage declines either to a "fixed" vcore setting, or something else.

It is possible therefore to set LLC to exceed VID under load -- I think that's what we're talkin' about.

With no experience yet on Ivy Bridge or Haswell, I would go into such a project (and I'm planning to . . ) with a presupposition that I want no more LLC than what it takes to leave a vDroop of a couple dozen millivolts under load.

Now . . . I suppose if this were "teaching," I'm using a textbook that's sort of like a tome on alchemy when you just want a basic up-to-date chemistry text. But I believe the basics apply here, so . . .

 

[rtsurfer]

There are unregistered or unmeasured spikes that occur when processor load ceases, and before voltage declines either to a "fixed" vcore setting, or something else.

It is possible therefore to set LLC to exceed VID under load -- I think that's what we're talkin' about.

With no experience yet on Ivy Bridge or Haswell, I would go into such a project (and I'm planning to . . ) with a presupposition that I want no more LLC than what it takes to leave a vDroop of a couple dozen millivolts under load.

Yes, Sir there is more up to date stuff out there.

You would think the technological world would have changed from 2007 to 2014.
It is freakkin 7 years.

Please man watch the damn Video .

There are 2 factors that can effect CPU voltage LLC and the Volt mode of the motherboard you are using. (e.g Manual, Adaptive, Offset).

Frankly, all of your posts on this page of this thread are annoying.
You are typing nonsense such as this

Again -- Ah been poh dis yeah -- my rental propitty been vacant. I'm still nursing this Sandy Bridge and happier-than-a-pig-in-s***. So I can only speak for this "Ol' Sandy Bridge" 2600K.

So you state some useful stuff but in between you fill it up with non-sense.

Yes we got in one post that you are still using a Sandy Bridge Processor.
No need to keep repeating that.

I was gonna report to the MODs but i couldnt think of a proper thing to state as a REASON.
Whatever..

 

[BonzaiDuck]

Yes, Sir there is more up to date stuff out there.

You would think the technological world wold have changed from 2007 to 2014.
It is freakkin 7 years.

Please man watch the damn Video .

There are 2 factors that can effect CPU voltage LLc and the Volt mode of the motherboard you are using. (e.g Manual, Adaptive, Offset).

Frankly, all of your posts on this page of this thread are annoying.
You are typing nonsense such as this



So you state some useful stuff but in between you fill it up with non-sense.

Yes we got in one post that you are still using a Sandy Bridge Processor.
No need to keep repeating that.

I was gonna report to the MODs but i couldnt think of a proper thing to state as a REASON.
Whatever..

[An attempt at humor, I suppose . . . I've been coming to these forums for 8 or 9 years. Some of the old hands would lapse into the "informal." Sorry that you're annoyed . . nothing more to say, I guess . . ]

MMmm. I couldn't find anything in the video that explains LLC. But it is "there" in the UEFI BIOS for the Maximus VI. The graphs in the "old" article would still provide an understanding of the topic . . Irrelevant that the paper deals with a six-year-old processor.

Another point. The video DOES address the issue of bCLK straps, and specifically recommends locking down the bCLK to 100. So you have to wonder what happens when trying to OC with that feature set on "Auto." You would think it would show up in CPU-Z and other monitoring software during an overclock attempt -- IF -- you can get the system to boot.

 

[rtsurfer]

[An attempt at humor, I suppose . . . I've been coming to these forums for 8 or 9 years. Some of the old hands would lapse into the "informal." Sorry that you're annoyed . . nothing more to say, I guess . . ]

MMmm. I couldn't find anything in the video that explains LLC. But it is "there" in the UEFI BIOS for the Maximus VI. The graphs in the "old" article would still provide an understanding of the topic . . Irrelevant that the paper deals with a six-year-old processor.

Another point. The video DOES address the issue of bCLK straps, and specifically recommends locking down the bCLK to 100. So you have to wonder what happens when trying to OC with that feature set on "Auto." You would think it would show up in CPU-Z and other monitoring software during an overclock attempt -- IF -- you can get the system to boot.

Well what you were writing wasn't funny to me. But HUMOR IS SUBJECTIVE, so we can throw that down the drain.
My apologies, I might have been a little bit too cranky as I have been having tough time in my life in the past two weeks.

Now back to topic.
The Video does deal with the issue with different modes like Manual , Adaptive, etc.
The problem OP is having is because he is stress testing in Offset mode instead of Manual.

If you let the BCLK to AUTO then it works most of the time. But sometimes under load it spikes up. Like last week I was testing a 47 multiplier and during testing LinX testing and my BCLK jumped from 100 to 104 and so I went from testing 4.7Ghz to 4.88Ghz which obviously resulted in a crash. First time it happened to me in months, so I went to manual BCLK.

And CPU-Z does monitor BCLK.

Alternatively use RealTemp Gt as it monitors your CPU clock and BCLK and also the temperature in the mix.


The final LLC issue, different motherboard have different LLC settings.
You have to play with them.
On my Asus Maximus Hero, there is an LLC setting from 1 to 8.
I keep mine at 6 because at 6 it is like .01V more than what I put in for the CPU Volt, but it is still good.
While at 5 it goes under what I set as the CPU voltage.
Some people just set that at max, I like to keep it where it gets closest to what I set it to.

 

[BonzaiDuck]

Well what you were writing wasn't funny to me. But HUMOR IS SUBJECTIVE, so we can throw that down the drain.
My apologies, I might have been a little bit too cranky as I have been having tough time in my life in the past two weeks.

Now back to topic.
The Video does deal with the issue with different modes like Manual , Adaptive, etc.
The problem OP is having is because he is stress testing in Offset mode instead of Manual.

If you let the BCLK to AUTO then it works most of the time. But sometimes under load it spikes up. Like last week I was testing a 47 multiplier and during testing LinX testing and my BCLK jump from 100 to 104 and so I went from testing 4.7Ghz to 4.88Ghz which obviously resulted in a crash. First time it happened to me in months but I still went to manual BCLK.

And CPU-Z does monitor BCLK.

Alternatively use RealTemp Gt as it monitors your CPU clock and BCLK and also the temperature in the mix.

We all have bad patches.

This "Offset versus Manual" issue had been bandied about over the last three years. A lot of people felt more comfortable simply fixing VCORE at a certain level and adjusting it for stability. What we found out with the 2nd gen chips was that you could choose "offset" mode, peg the offset at a +/- increment, then adjust the "Extra Voltage for Turbo." It may be for different chips, the best increment would be different from another.

But we're two generations of chips along from that. While it would pay to understand what offset and the "extra" setting do (or did) with regard to VID, it would also add complexity. So I won't take a stand on this.

I was candid that I don't have an IB or Haswell to play with at the moment. But I've been observing that there are possibly some to whom this is fairly new. there may be others who -- like me -- have conceptions based on earlier processor lines. Conceptions are fine, but it can be easy to make assumptions, or forget about certain factors in the BIOS -- as for instance the bCLK and how it's no longer fixed by default at 100 -- that there are different "straps" that were added when they released SB-E.

Personally, I'm not sure what to make of Haswell. People who don't have the troubles discussed here nevertheless find it limited to something like 4.4Ghz. And I can't remember if I saw someone here who had reached 4.6 or 4.7. Don't think I did, unless that's what you've done. The IB cores seemed more in line with the generation before, and for that -- more overclockable, except for the TIM problem.

Even the video seemed to suggest as much. If the chip couldn't be run out of spec, you got a bad one and tough luck.

But I'd seen another article pertaining to Haswell, the die-size, the cooling problem -- suggesting in the title and then as the article's conclusion that there would be a "slow death of overclocking."

That's why I may build a new system this year, but it may be an IB-E core.

 

[JimmiG]

I successfully delidded my 4770k with decent results when comparing the same settings. Before my CPU would max out at 78 C on a very high IBT 10 run loop @ 4.0ghz 1.09v. Now I'm hitting 61 C at the same settings. So thats a 17 C drop and that seems to be "average" for successful delids.

As I try to push my chip moderately higher, though, I'm having to massively increase the voltage to get any gains at all in core speed. For a measly 200mhz increase (up to 4.2ghz), I need to go from 1.09 to 1.162 volts, and temps are back into the low 70's. I'm using a Thermaltake Frio with a single 120mm fan so I know it's not going to get out-of-this-world results, but for another 200mhz that seems rather ludicrous to have a 10 degree temperature spike and that much of a voltage increase when considering how modest of an overclock it is to begin with. Is this fairly common with 4770k overclocking? I know there are a wide range of "good" and "bad" chips out there when it comes to Haswell, but I thought for sure I'd be able to get to 4.4ghz and keep temps at or below 70 C.

A delid buys you anywhere from 10C to 30C, but a fairly small increase in Vcore will quickly eat into that additional temperature headroom. Delidding is not a miracle cure. If your Haswell needs a lot of additional voltage for every 100 MHz, you'll still hit a temp brickwall.

However your temps and voltages are very conservative. Don't worry about using 1.25V - 1.30V as long as the temp is below mid 90's C.

Example with my Haswell:
Before de-lid: 4.3 GHz / 1.218V
After de-lid: 4.5 GHz Adaptive (goes to about 1.35 - 1.36V running stress tests)

 

[BonzaiDuck]

A delid buys you anywhere from 10C to 30C, but a fairly small increase in Vcore will quickly eat into that additional temperature headroom. Delidding is not a miracle cure. If your Haswell needs a lot of additional voltage for every 100 MHz, you'll still hit a temp brickwall.

However your temps and voltages are very conservative. Don't worry about using 1.25V - 1.30V as long as the temp is below mid 90's C.

Example with my Haswell:
Before de-lid: 4.3 GHz / 1.218V
After de-lid: 4.5 GHz Adaptive (goes to about 1.35 - 1.36V running stress tests)

Is that the statistical range of your monitored or recorded voltages while your stress test is running -- minimum to maximum?

 

[stuff_me_good]

Worth delidding?


My 4770k goes 4.3Ghz @ 1.25v ibt and prime stable, but the temps are pretty high already. I think it would go even higher clocks, but the temps become unusable high. I cant even try to run 1.3v @ 4.5ghz because the temps go 100C almost immediately after initiating IBT. In windows 4.5@1.25v runs fine(have not gamed or done anything cpu intensive though).

Also there is very big difference between the core temps. Two of them are running 10-8C higher than the other two.

So it seems that even though the processor could clock reasonable high but I even can't try because the temps just blow through the roof if I try a little bit.

I have phanteks TC14PE and liquid pro between the cpu and the cooler.

 

[JimmiG]

Is that the statistical range of your monitored or recorded voltages while your stress test is running -- minimum to maximum?

At 4.3 GHz I was using a fixed voltage of 1.218V. At 4.5 GHz I'm using Adaptive. It tops out at 1.265V running regular games and stress tests, but AVX kicks it into the 1.36V range.

Since I don't normally run AVX stress tests all day, I figured it's better for the CPU to use Adaptive, as this allows the VCore to drop when the CPU is idle.