Intel Haswell Overclocking Clubhouse.

[DOM]

I think I got the worst rma chip ever 1.6v for 5GHz @-55c

 

[Deleted member 24505]

I found this, is interesting.

"Now, I’ll share a secret imparted by the folks at ASUS who gave several reviewers some tips on overclocking the retail stepping Haswell chips: Set Vcore to 1.20 V. Set all cores to 46x (which would be a 4.6 GHz overclock), save & reboot. If the system boots past the UEFI and either begins to load or, ideally, makes it into the OS and is stable, you have a 50th percentile or greater chip on the Haswell overclocking-ability bell curve. If it won’t at least boot there and make it into the UEFI, you probably have less than a 50th percentile chip. You can expect chips in the lower 50th percentile to top out in the 4.4-4.5 GHz range at 1.25 V.

If your chip will boot at 4.6 GHz and 1.25 V, that’s very good. It means you have at least an average chip. If it will boot at 4.6 GHz and is stable there, then you may have an above average chip. The best chips will be able to do 4.8 GHz stable at 1.25 V. Our sample did 4.8 GHz, but at 1.3 V and on a custom water loop. Using 1.3 V will likely put a chip out of the air cooling / AIO water cooling thermal envelope. Temperatures in all of these scenarios, from the dog 4.3 GHz chips up to the good 4.8 GHz chips, will always be in the ~90°C range. That’s just the nature of Haswell. With the VRM on-die, think of Haswell as Ivy Bridge plus 10° C."

From here-
http://www.overclockers.com/3step-guide-to-overclock-intel-haswell

 

[cadaveca]

I found this, is interesting.

"Now, I’ll share a secret imparted by the folks at ASUS who gave several reviewers some tips on overclocking the retail stepping Haswell chips: Set Vcore to 1.20 V. Set all cores to 46x (which would be a 4.6 GHz overclock), save & reboot. If the system boots past the UEFI and either begins to load or, ideally, makes it into the OS and is stable, you have a 50th percentile or greater chip on the Haswell overclocking-ability bell curve. If it won’t at least boot there and make it into the UEFI, you probably have less than a 50th percentile chip. You can expect chips in the lower 50th percentile to top out in the 4.4-4.5 GHz range at 1.25 V.

If your chip will boot at 4.6 GHz and 1.25 V, that’s very good. It means you have at least an average chip. If it will boot at 4.6 GHz and is stable there, then you may have an above average chip. The best chips will be able to do 4.8 GHz stable at 1.25 V. Our sample did 4.8 GHz, but at 1.3 V and on a custom water loop. Using 1.3 V will likely put a chip out of the air cooling / AIO water cooling thermal envelope. Temperatures in all of these scenarios, from the dog 4.3 GHz chips up to the good 4.8 GHz chips, will always be in the ~90°C range. That’s just the nature of Haswell. With the VRM on-die, think of Haswell as Ivy Bridge plus 10° C."

From here-
http://www.overclockers.com/3step-guide-to-overclock-intel-haswell

Ask W1zz if I didn't tell him exactly this little "ASUS secret" prior to the launch, when he got his chip. I had chips before he did. For all I know, he told that to ASUS and ASUS ran with it. I might have even had retail chips before ASUS did, and if you know a bit of the backstory about Haswell and ASUS, it might even make sense. Anyway, that's been the general rule since Ivybridge. Since these chips are just IvyBridge with added stuff, bigger IGP, AVX2, and FIVR, naturally the same rules apply to clocking the CPU portion of the chips. The thing is, you can know how good of a chip you have by simply booting the first time and checking the BIOS for stock voltage. The rest of that is all just general BS.


The idea that it's IVB +10c, and that expectation, might be why some people seem to have issues, although most of the guys here in this thread seem to be doing relatively fine for 24/7. Haswell is simply way more than just that.

 

[Deleted member 24505]

Ask W1zz if I didn't tell him exactly this little "ASUS secret" prior to the launch, when he got his chip. I had chips before he did. For all I know, he told that to ASUS and ASUS ran with it. I might have even had retail chips before ASUS did, and if you know a bit of the backstory about Haswell and ASUS, it might even make sense. Anyway, that's been the general rule since Ivybridge. Since these chips are just IvyBridge with added stuff, bigger IGP, AVX2, and FIVR, naturally the same rules apply to clocking the CPU portion of the chips. The thing is, you can know how good of a chip you have by simply booting the first time and checking the BIOS for stock voltage. The rest of that is all just general BS.


The idea that it's IVB +10c, and that expectation, might be why some people seem to have issues, although most of the guys here in this thread seem to be doing relatively fine for 24/7. Haswell is simply way more than just that.

Did you read any of this guys stuff in this guide? what do you think to it, or just the usual stuff.

did you know about this Asus trick before then? were did you hear of it? does it give a good indication of the capabilities of the chip?

Sorry for all the questions, I like to take in all the stuff i can.

tigger

 

[cadaveca]

Did you read any of this guys stuff in this guide? what do you think to it, or just the usual stuff.

did you know about this Asus trick before then? were did you hear of it? does it give a good indication of the capabilities of the chip?

Sorry for all the questions, I like to take in all the stuff i can.

tigger

I read the full guide. It's based on an ASUS guide, which I have as well. That guide says this:

1. Approximately 70% of CPUs can go up to 4.5GHz. Overall, most CPUs are capable of
reaching 44x to 45x with varying levels of voltage. Voltage will be a key item as it defines thermal
output. It is possible to run out of thermal headroom at lower frequencies due to a processor
requiring excessive voltage for stability.
2. Approximately 30% of CPUs can go up to 4.6GHz
3. Approximately 20% of CPUs can go up to 4.7GHz
4. Approximately 10% of CPUs can go up to 4.8GHz
Currently we are seeing voltage ranges between 1.150v to as much as 1.400 for similar/same
frequencies. An example would be 4.6GHz requiring 1.200 and another CPU requiring 1.400.
This is important as a the overall maximum load a single radiator single 120mm closed loop
cooling solution can dissipate under synthetic stress test load is approximately 1.250 to 1.275v.
This is assuming the ambient temperature is reasonable.
A voltage level of 1.350v or greater, is too high for enthusiast cooling solutions including water
cooling. With this in mind, maximum temperatures should be gauged by the use of real world
applications rather than synthetic stress tests.
Given Haswell’s thermal characteristics, pursuing per core overclocking can be beneficial as this
helps to maximize core frequency for various loading conditions.

I had the "trick" idea before ASUS gave out their guide. I came up with it myself after testing a few chips, retail and ES, and W1zzard asked me what I thought and at that moment, that's what I told him. Like literally, those exact words, set all cores to 46, set voltage to 1.2V, boot into Windows. But like I said, that's what I used for IVB, and expected the same.

And now, no, I do NOT think it gives a good indication of chip capabilities, as what you can boot and what is required for 24/7 stability are not so close and depends on many things.

Noone seemed to read this part, however:

It is advised once exceeding 46x multipliers to consider Per Core overclocking. This can provide
higher frequency performance relative to threaded loads, but offers the advantage of not
requiring additional VID or producing additional heat.
An example is
3 and 4 cores – 47x
1 and 2 cores – 48x

 

[Deleted member 24505]

I read the full guide. It's based on an ASUS guide, which I have as well. That guide says this:





I had the "trick" idea before ASUS gave out their guide. I came up with it myself after testing a few chips, retail and ES, and W1zzard asked me what I thought and at that moment, that's what I told him. Like literally, those exact words, set all cores to 46, set voltage to 1.2V, boot into Windows. But like I said, that's what I used for IVB, and expected the same.

And now, no, I do NOT think it gives a good indication of chip capabilities, as what you can boot and what is required for 24/7 stability are not so close and depends on many things.

I guess each chip is different in some way, so each owner has to find their chips foibles.

 

[cadaveca]

I guess each chip is different in some way, so each owner has to find their chips foibles.

Yeah, but thankfully, the introduction of all these separate domains has made it less of an issue, it just takes more time tweaking to find the sweet-spot for everything together.

As to ASUS's Voltage considerations:

Quick note regarding voltage scaling – VID for frequencies varies considerably from CPU to
CPU. It is strongly advised offset/adaptive values be used for superior efficiency, as reducing
chances of voltage degradation.
Voltage variance between samples: outstanding CPUs at 4.6GHz require 1.150 VID and other
CPUs requiring 1.400 VID for 4.6GHz.
The voltage recommendations below target higher quality CPUs. If the settings work, you will be
able to POST and BOOT without issues, as well as run initial stability tests like AIDA Benchmark
Report.
CPUs that are lower in their scaling capability and require more VID consistently fail to BOOT the
OS volume.
Enter Ai Tweaker
Define Manual Operation – Do not initially define XMP. This is to ensure that the DRAM divider
does not impact CPU frequency scaling. This should be left at a default JEDEC value of 1333 or
1600.
Select Core Ratio Limit – define 46x
Do not initially define a CPU Cache Ratio – allow the board to define the AUTO value. This will
help to ensure a smooth OC experience. Once you have validated the stability of a 4.6GHz
overclock, you can go back and manually define a Cache Ratio in parity with the CPU ratio. This
will provide the best IPC/performance. Be advised not all CPUs will be able hold a native Cache
Ratio.
Digi+ Power Control –
All options can be maintained at AUTO values.
CPU Core Voltage
There are 3 options
Manual – Override V/f
This is a static VID and offers the easiest level of understanding in how it works in conjunction
with the CPU and VRM. It offers the worst efficiency and produces the highest power
consumption as well as will produces higher idle and moderate load temperatures. It also has a
higher likelihood of creating processor voltage degradation over time.
The main advantage is it will not be impacted by synthetic stress test applications like Prime 95,
Aida64, OCCT or others. For the highest level of overclocks, this type of VID may aid in stability
as you do not want movement in the voltage being supply at higher clock speeds.
Overall, in this mode all other inputs are ignored and the voltage is driven at the specific value
across all frequencies.
Offset – Offset V/f
This mode is a flexible and variable VID. It offers superior efficiency compared to a manual VID.
Offset yields improvements in power consumption and thermals over static VID. In this mode the
CPU voltage will vary based on frequency.
A disadvantage is it will apply an offset even to the base CPU VID when the processor is idle.
This mode applies an offset, either positive or negative, to the entire voltage stack.
Adaptive – Interpolation V/f
The mode is the most flexible and variable VID. It offers the greatest efficiency. It yields the best
power consumption and thermals. It will vary depending on the CPU frequency.
Adaptive mode offers the advantage of not impacting the non Turbo Ratio based multipliers.
Essentially all standard VRD voltages will be maintained at stock levels. Additional voltage will
only be supplied for greater than standard CPU ratios / multipliers.
Adaptive VID is only applicable for frequencies and voltages beyond the default maximum turbo
frequency. The UEFI can be set to request an additional amount of voltage that should be
applied when running at the highest requested overclock CPU multiplier ratio.
For a 4.6 GHz overclock attempt: However you set the voltage, it is advised you set a VID of
1.200. This will be for great quality CPUs, outstanding quality CPUs will be 1.175 and the best
CPUs will be 1.150. *
* This is assuming the margin of the CPU allows for this multiplier to even be reached. Most CPUs will
have a limited range based on margin, and voltage required for corresponding frequency.
For a 4.8 GHz overclock attempt: However you set the voltage, it is advised you set a VID of
1.300. This will be for great quality CPUs, outstanding quality CPUs will be 1.275 and the best
CPUs will be 1.275 and below. For the majority of CPUs a voltage range of 1.325 to 1.400 will be
required for 4.8GHz.*
* This is assuming the margin of the CPU allows for this multiplier to even be reached. Most CPUs will
have a limited range based on margin, and voltage required for corresponding frequency.
Cooling Recommendations
Recommendations are based on voltages, as variance in CPU frequency makes targeting
frequency inconsistent.
For voltages up to 1.250-1.265 a cooling solution meeting a minimum of a Corsair H80i is
advised. Superior performance can be offered by moving to the H90 or higher performing dual fan
closed loop solutions.
For voltages up to 1.275-1.300 a cooling solution meeting a minimum of a Corsair H100i is
advised.
Keep in mind these recommendations are based on controlling peak temperatures and loads
under synthetic stress applications. For nominal real world usage, load temperatures will be
considerably lower than stress tests.
For voltages up to or greater than 1.300v a high performance water cooling system is
recommended. Minimum recommendation would be a unit such as Koolance EX2-755.
For voltages up to or greater than 1.350 a high performance water cooling system is
recommended. Minimum recommendation would be a unit such as a Koolance EX2-1055

So, because this is all from ASUS's guide, and what overclocker's reports is a shortened version of that with their own input, there's some good info there, but at the same time, this was a guide written for reviewers when benchmarking, not really working for 24/7 clocking. This is just a guide based on getting results for reviews. In the end, my own guide isn't much different, honestly.

 

[Deleted member 24505]

I noticed he said clock speed is king, in a choice twixt core speed and ram speed, take the higher core speed, is this true for Haswell. He also said low latency on timings is better.

 

[cadaveca]

I noticed he said clock speed is king, in a choice twixt core speed and ram speed, take the higher core speed, is this true for Haswell. He also said low latency on timings is better.

I guess he's referring to the fact that CPU core speed should be first priority when reaching for performance. IF anything you adjust affects CPU speed, you might want to reconsider that change that affected the CPU cores, in other words.


The rest of the guide, I don't agree with, really. It's more focused on OC-benching for extreme guys, not for 24/7 use. Perfect example is needing 1.15 V to 1.25V for VCCSA...you really shouldn't need that for 2933 MHz. I have 2933 MHz and 3100 MHz DIMMs, neither require adjust to VCCSA unless using BCLK adjustments. Memory...timings vs speed depends on the type of stick. Remember, that's a guide for benchmarking for reviews.

I could post the full guide, no problem, but I think that since you can get it elsewhere, there is no point.

 

[Jstn7477]

Hey Dave, I brought home my Gigabyte 7950 Windforce and when inserted into my ASRock Z87 Extreme6, I get a b2 post code. A dying XFX 7950 with the same ROM worked fine in the same slot with an MSI 7970 as a slave. Any idea what could be wrong?

 

[cadaveca]

b2 is just before BIOS, no idea, could just be a bad card since BIOS should be just about to show screen then.

 

[Jstn7477]

Weird, it worked fine in my Z77 Extreme6 just an hour ago. This sucks...

 

[Kursah]

*sigh*

Well I thought I had 4.3 stable at 1.26v...got 124 code in Wargames EE...tried 1.27 and 1.28v...again bsod 124. It's odd...it just seems at a random spot during the day the chip just gets unstable. I'm annoyed and back at stock speeds atm so I can at least game for a bit without the damn thing crashing. I all but gave up on 4.4...just too much voltage and heat for my air cooling to handle this time of year...better luck come winter. This chip is more and more disappointing every time I try to find stability and usability.

I've cleared bios, reflashed 1205. I noticed that the cpu vrm would stick at 1.712 no matter what I set once OC-ing..not sure if this is normal or not...or if it even matters. Works ok on auto and stock.

 

[cadaveca]

Weird, it worked fine in my Z77 Extreme6 just an hour ago. This sucks...

Is ya sure the IGP is disabled? try another slot. can't say I've experienced anything like that, honestly.

 

[Jstn7477]

Swapped the 7970 to slot 1 and GB 7950 to slot 2 and it posts now. Time to test and make sure this card works fine in CFX unlike my XFX 7950 that now throttles and freezes with any load applied to it besides Aero.

 

[HammerON]

I have been running @ 4.4GHz for the last three weeks while crunching. No BSODs:

 

[petedread]

*sigh*

Well I thought I had 4.3 stable at 1.26v...got 124 code in Wargames EE...tried 1.27 and 1.28v...again bsod 124. It's odd...it just seems at a random spot during the day the chip just gets unstable. I'm annoyed and back at stock speeds atm so I can at least game for a bit without the damn thing crashing. I all but gave up on 4.4...just too much voltage and heat for my air cooling to handle this time of year...better luck come winter. This chip is more and more disappointing every time I try to find stability and usability.

I've cleared bios, reflashed 1205. I noticed that the cpu vrm would stick at 1.712 no matter what I set once OC-ing..not sure if this is normal or not...or if it even matters. Works ok on auto and stock.

Just when I had come to terms with having to live with 4.3 (@1.219v), I've just ran prime95, blend test, I'm getting errors. Even with ram at 1600. Gone back to default bios settings and prime95 is running fine. Wow what a chip.

(Edit) luckily I've made a mistake, earlier today I had to go out half way through experimenting with lowering the VRIN. I had forgotten that I had left it at 1.850v when it needs 1.860v. So it looks like I can have my 4.3 oc.
Seems like you guys on Asus boards don't need anything like 1.8## even for higher multi's.
Now I can get back to trying to get my ram to 2400mhz.

I'm sure I have the latest version of prime95.
Aida64 ran for 20 hours and OCCT I ran for 3 hours, but prime95 workers fail with my tiny little 4.3 OC.

It's not like I can sell this chip. Who in their right mind would by a 4770K from Ebay.

 

[TheHunter]

*sigh*

Well I thought I had 4.3 stable at 1.26v...got 124 code in Wargames EE...tried 1.27 and 1.28v...again bsod 124. It's odd...it just seems at a random spot during the day the chip just gets unstable. I'm annoyed and back at stock speeds atm so I can at least game for a bit without the damn thing crashing. I all but gave up on 4.4...just too much voltage and heat for my air cooling to handle this time of year...better luck come winter. This chip is more and more disappointing every time I try to find stability and usability.

I've cleared bios, reflashed 1205. I noticed that the cpu vrm would stick at 1.712 no matter what I set once OC-ing..not sure if this is normal or not...or if it even matters. Works ok on auto and stock.

Did you tweak cache multi and its voltage?

Well I got the same thing today, I ran realtemp 1.0rc and cinebench11.5 to see the temps with new noctua FLX A14 push/pull fans and baam 0x124 wmea or what ever bsod..

I went to uefi set min cache to 39x instead of 8x, max 41x adaptive offset 1.13v and Cinebench11.5 passed normally. Now it idles at 1.060v.



I also tweaked some windows bcdedit commands, one is specifically for faster cpu sleep to wake response >> bcdedit /set disabledynamictick yes, I had it enabled and now its deleted and set to sys default.
Idk maybe balanced power plan with default value didnt react fast enough, but it could be just big coincidence.

 

[Kursah]

Did you tweak cache multi and its voltage?

Well I got the same thing today, I ran realtemp 1.0rc and cinebench11.5 to see the temps with new noctua FLX A14 push/pull fans and baam 0x124 wmea or what ever bsod..

I went to uefi set min cache to 39x instead of 8x, max 41x adaptive offset 1.13v and Cinebench11.5 passed normally. Now it idles at 1.060v.



I also tweaked some windows bcdedit commands, one is specifically for faster cpu sleep to wake response >> bcdedit /set disabledynamictick yes, I had it enabled and now its deleted and set to sys default.
Idk maybe balanced power plan with default value didnt react fast enough, but it could be just big coincidence.

Ya I have messed with cache multi, varied min/max, set to auto, adjusted voltage,etc.

I am back to OCing again today...I just went back to stock last night and let it run that way for the hour or so I gamed before going to bed. I ran Intel's ETU at a revised and semi-stable 1.260v to find it has BSOD'd 124 and restarted sometime during the night..I had the test set for 1h 5m so within that hour and five minutes it crashed. I'm at 1.265v and it reads 1.267 under load and so far it's okay for now...

My chip in Adaptive idles with .699v-.709v, the stock vid is 1.071v iirc...so not a great chip but I'm hoping that I can just mayyybe find some stability.

Maybe I will try setting my min cache to 39 or auto again and see what happens...I just assumed it would be a good idea if it was clock switching too...but it's not like my chip runs noticeably cooler with it clocking down during idle speeds. I do have a better 120mm exhuast fan coming for my case and my Noctua NH-U14S has been a champ...love the cooler!

One thing I noticed is that my VRM wouldn't deviate form 1.712-1.728v yesterday. No matter what I set it to. If I put everything back to stock/auto it would read 1.798v. Right now I have it set to 1.8v and it's holding with a reading of 1.808-1.828v. So far things are pretty stable...but then again I have't given it the gaming gauntlet yet.

 

[TheHunter]

^
nvm that small SVID variation, if you want it fixed then set it to enabled and put in 1.75v at around 1.25-1.30v area (just in case).

btw if disabled its at 1.70v

-----

A bit off topic, but does anyone know what's better for push/pull fans
intake from outside - in my case rear exhaust, or as intake inside the case?

Well I had one fan as intake push from rear exhaust and it cooled just as good like now push/pull as intake inside the case.. I have another 200mm exhaust on top and one 20mm by side as intake, so heat buildup isnt a issue.


Although I saw another recommendation; two push fans on top getting air from outside as intake inside, apparently the best scenario.

 

[Kursah]

Gotcha...I need to check but I'm pretty sure I have my SVID disabled and set to 1.8v and it's reading 1.808v atm. I may drop it to 1.75v, and enable it later on to see what happens.

I had to help my sister out with a flat tire on her Jeep so I let my PC run the Intel ETU CPU test for a couple hours with a 1.265v CPU voltage and so far it's stable...too bad I need so much voltage just for 4.3Ghz but maybe having the even higher CPU VRM voltage is necessary in my case. We'll see...gonna play some Wargames: EE and see what happens.

 

[Womper]

Yeah, these long-term BSOD are challenging. I'm going to throw voltage at it to see if it goes away. Instead of 1.26v adaptive, I bumped up to 1.3v. I also disabled SVID control and set input voltage to 1.8.

It is advised once exceeding 46x multipliers to consider Per Core overclocking. This can provide
higher frequency performance relative to threaded loads, but offers the advantage of not
requiring additional VID or producing additional heat.
An example is
3 and 4 cores – 47x
1 and 2 cores – 48x

I tried per-core mode a while back, and found it unhelpful. At the same voltage, sync mode at 47x was fine but per-core with 47x on all cores bluescreened quickly. Even something like 47x, 46x, 45x, 44x was unstable at my 47x voltage...so I didn't see any way to save core voltage. Maybe I'll try again.

 

[Kursah]

I have only tried linked cpu multipliers thus far. I'd like to think that maybe a couple of my cooler running cores might be capable of a 44x when the hot ones are at 43x...but that's probably pipe dreams. But after a couple hours of gaming, several hours of stability testing it seems 1.265CPUv, 1.135 CacheV (still testing...) and 1.8v SVID is stable for my chip's 4.3GHz operation. For now at least!

Makes me wanna try for 4.4 again lol! Damn I wish I had money to grab another one of these and hope for a better performer. Though I gotta say when this chip is running stable the performance is great!

 

[TheHunter]

My SVID is at auto atm, at first it was ~ 1.71v - 1.75v, now its stuck at 1.79v.

And I didnt have any issues for 2 days. I guess it really needs ~ 1.75-1.85v at higher voltage


Edit: yes, tnx Dave

Check this intel pic by Vccin it says 1.80v is nominal

:

 

[cadaveca]

My SVID is at auto atm, at first it was ~ 1.71v - 1.75v, now its stuck at 1.79v.

And I didnt have any issues for 2 days. I guess it really needs ~ 1.75-1.85v at higher voltage


Edit: yes, tnx Dave

Check this intel pic by Vccin it says 1.80v is nominal

But the Whitepapers published after that pic say differently. I was me that posted that pic, after all, so yes, I am aware of what it says. it also says nominal up to 2.3 V, and static up to 3.04V, which is kinda of true, but can and will kill chips.

There's no mystery about this for me really, this is published information.