Intel Haswell Overclocking Clubhouse.

[Womper]

EDIT Part Deux: Well in that time (what maybe 30 minutes?) Wargames: EE caused a 0x124 BSOD....so while AIDA was more stable...gaming sure isn't...

Intel Extreme Tuner's stress test seems to be roughly the same as game stress, in the sense that it will bluescreen under the same configurations in about the same amount of time. AIDA seemed to take longer.

If you're just trying to game, I would assume you have plenty of thermal headroom. For this sort of usage, I think you should just switch to a 1.32v - 1.35v static voltage for your core. That should work for 4.4GHz. If heat still isn't an issue, I'd take it to 1.37v and shoot for 4.5GHz.

As for the cache, I'd go with 8x min 39x or 40x max, and 1.3v adaptive- just to remove it as a variable while you tweak your core up.

 

[TheHunter]

I've been testing 4.7ghz @ 1.255v a bit and it looks quite promising,

cache 35x, 41x @ 1.118v adaptive offset
auto SVID 1.79v
LLC6
120%




3dmark vantage 35k vs i7 2600k @5.1ghz 33.8k or i7 3770k @4.8 33.4k

but its not fully stable though, 1.255v failed in FFXIV reborn benchmark.. Will try some more tomorrow, imo 1.26v should do the trick and pass the rest

 

[Kursah]

Well I've been reading forums and finding that people are playing with adding .050+v to their system agent, vccio settings to help smooth out stability with higher than 1600 ram speeds and also some have found that increasing a little voltage here can also allow "some" folks to lower their cpu vCore. I forget the forum I read it but a couple guys claimed that the 0x124 DTC's weren't necessarily from too low of vCore anymore that there is so much more at play that mess with stability and that DTC being set.

So I'm gonna mess with that in due time. Had a storm come through and knock out power for about 12 hours and now I gotta honey-do list to conquer yet today. But so far I'm running Intel's test for now...

On running my Cache at lower than 1.19-1.20v I might try again, but I have found some stability going above the 1.16v mark at least a hopeful placebo lol. But I had it running in Adaptive, and I find no point in this setting for cache because it just locks the voltage at that setting...I can do that in manual voltage setting too. But Auto and Offset both allow the voltage to adjust with cache clocks in a power state, like the CPU. Whether or not this has allowed me more stable operation at one point I would say ya...but with the touch and go of this chip I don't know...I need to start keeping a more thorough Haswell OC journal lol! Maybe I should go back to Adaptive on the cache? Heck I dunno!

 

[cadaveca]

a couple guys claimed that the 0x124 DTC's weren't necessarily from too low of vCore anymore that there is so much more at play that mess with stability and that DTC being set.

That is correct, memory unstable can make 0x124 BSOD, too.


One of the regulars in our TS, MXPhenom, he got the HERO and some 2133 MHz 1.5 V Dominator Platinums, got 0x124. Set looser tFAW, now fully stable.


Crazyeyesreaper was getting 0x124...for him, we disabled IGP.

VCCSA should be 0.85V or so @ stock, I run 1.05V on VCCSA, VCCAIO and VCCDIO. DIO and AIO are 1.02V stock for me, so just a minor boost.

 

[Kursah]

Well I relaxed things a bit, started over again with a lowered Cache speed at 8-35x and ram back at 1600 for now...got 4.4 testing stable with Intel's ETU @ 1.35v. That just seems like so much voltage..but my temps are not going beyond 90C yet...though I am trying to keep the house cooler. That and I haven't ran AIDA's cpu stress test yet... I figure if it can pass an hour of Intel, then move onto AIDA...then when I get home later try Wargame.

I suppose it makes sense that memory can cause a 124 code too with the memory control being integrated into the CPU. I have NOT disabled my IGP. Maybe I should...I just assumed not use it, but leave it enabled should the day come my GTX 770 crap out or something. But I also suppose a clear cmos would be just as beneficial to getting it back if I don't have a backup card again by then (every time I get one, someone needs it! lol last one was a GTS250 I got for like 45 bucks..kid's gaming card smoked the day I got it! lol)


Well I will keep messing around and report back. I just wish there was a 5-minute or even 15 minute test we could run. I guess 15 minutes of AIDA64 should help with initial stability. That and a similar session in Wargame would really let me know pretty quick in all honesty I suppose.

I may go back to trying to find 4.3Ghz in the sub 1.30v range...just not sure I'm comfortable running 1.35v...

 

[Womper]

That is correct, memory unstable can make 0x124 BSOD, too.


One of the regulars in our TS, MXPhenom, he got the HERO and some 2133 MHz 1.5 V Dominator Platinums, got 0x124. Set looser tFAW, now fully stable.


Crazyeyesreaper was getting 0x124...for him, we disabled IGP.

VCCSA should be 0.85V or so @ stock, I run 1.05V on VCCSA, VCCAIO and VCCDIO. DIO and AIO are 1.02V stock for me, so just a minor boost.

I have the CL9 2133 1.5v dominator platinums, maybe I should do that. What setting did he end up using?

 

[cadaveca]

I have the CL9 2133 1.5v dominator platinums, maybe I should do that. What setting did he end up using?

Does your BIOS have SPD tool in tools section?

 

[MxPhenom 216]

I will post pictures tonight, but only thing I had to change to get mine working is the jFAW from 32 to 40.

 

[Kursah]

Man..well I actually had an easier time tonight with 4.4 being stable than 4.3. Especially in gaming! I had a solid hour of Wargame crash free and overclocked! WOOT! Hell ya! You guys have NOO Idea! But at the same time it makes me want to RMA my chip in the hopes of one less shitty than this one.

But now that things have cooled off now for some reason instability started happening. Well I did eventually x124 outta the game, and after a little over an hour of AIDA testing after that. Then from there it went to shit. Well my SVID locked at 1.7v, which reads 1.712v. No matter if I set to Auto, enabled, or disabled. This has been an issue before where it locks at that..usually an F10 on Auto SVID and Auto Voltage Variable clears it up. Gonna have to clear CMOS this time. Anyone else notice this? I know someone mentioned earlier when SVID is set to Disabled it defaults to 1.7v...that just isn't true...it's a bug/glitch. Because I set to disabled many times before and a couple days since my last CMOS clear and was able to change anywhere between a tested 1.65v and 1.90v with readings in BIOS and AIDA corresponding the change. They both also correspond to the voltage lock.

Just figured I'd share another issue that is causing me instability concerns with the crowd, get some feedback and see what ya'all are doing about it.



Edit: Cleared CMOS, SVID is now adjustable again. Now to see if I can't keep this thing from crashing on every stability test above stock! lol!

 

[Womper]

Does your BIOS have SPD tool in tools section?

Yeah, it shows the second and third timings on the DRAM page with the main timings. Second and third are all on Auto and it shows their values...I assume I can adjust them.

I will post pictures tonight, but only thing I had to change to get mine working is the jFAW from 32 to 40.

So what kind of BSOD issue were you fighting? Or were you able to get stability at a lower voltage? I'm having a tough time getting Adaptive mode to be stable on the core. I have a 4.8GHz config with a 1.34v static voltage (4.5GHz cache @ 1.3v adaptive, 2133 ddr)...not a single crash yet. If I change that to a 1.34 adaptive (no other changes), or even a 1.35 or 1.36v adaptive, then I bluescreen 30min-1hr into a game or IET stress test.

Well my SVID locked at 1.7v, which reads 1.712v.
Edit: Cleared CMOS, SVID is now adjustable again. Now to see if I can't keep this thing from crashing on every stability test above stock! lol!

That sounds really low for VRin. I can't remember- what's your normal setting? I'd go with 1.8v with SVID disabled.

 

[MxPhenom 216]

Yeah, it shows the second and third timings on the DRAM page with the main timings. Second and third are all on Auto and it shows their values...I assume I can adjust them.



So what kind of BSOD issue were you fighting? Or were you able to get stability at a lower voltage? I'm having a tough time getting Adaptive mode to be stable on the core. I have a 4.8GHz config with a 1.34v static voltage (4.5GHz cache @ 1.3v adaptive, 2133 ddr)...not a single crash yet. If I change that to a 1.34 adaptive (no other changes), or even a 1.35 or 1.36v adaptive, then I bluescreen 30min-1hr into a game or IET stress test.



That sounds really low for VRin. I can't remember- what's your normal setting? I'd go with 1.8v with SVID disabled.

I have not done any CPU clocking yet. Waiting for Dave's guide to release and do a thorough read over of it.

Dave, like my 2500k build, helped me get 2133 working by changing a few timings.

I am running fully stable with 2133 at 1.5v. XMP on, but my board/bios set one timing, the jFAW too tight so it would freeze and bluescreen with 0x124 code about 20 seconds after hitting the desktop. Changed the jFAW timing from 32 to 40 which is what it should be been set too based on the DIMMs programming, but the board set it to 32.

Once switch to 40, and everything else kept as is, things have worked flawlessly.

 

[dumo]

3600 single air 32m



Ram on cold valid

 

[Kursah]

That sounds really low for VRin. I can't remember- what's your normal setting? I'd go with 1.8v with SVID disabled.

That's what it defaults to when I get so many crashes I believe. Then cleared CMOS default is 1.79 in BIOS so I'm assuming 1.80v. I have it manually set at 1.80v with SVID to Disable. But that's where it was prior...just after so many crashes it seems to lock itself to 1.70v for some odd reason.

Had some luck with 4.3 @ 1.26v, 8-39X Cache @ 1.15v, Agent at + .100v, SVID Disabled / 1.80v, Memory at 2133 XMP, 1.60v (set to 1.58v in BIOS) 11-11-11-28 cr1. It ran Wargame for a couple hours while I watched a movie. Is an hour into a several hour test with Intel's ETU on CPU Test, will do Memory test next. Interesting how it's so stable at 1.26v where it was so spotty before...the System Agent at +.100v is to thank for that at this point, and the SVID being adjusted to 1.80v I believe has a big part to play as well.

Still working on fine tuning...4.3 and 4.4 settings, will also see what 4.2 takes...under stock load I can see upwards of 1.21v iirc...see how far I can get which I'm assuming is 4.2 maybe. It's too bad you can't adjust multi's by the tenth...I'm sure in the next generation or two that will happen... maybe I could hit 43.3X Multi. Of course I suppose if a person wants to mix BCLK and multi's that can be achieved already.

 

[Womper]

I have not done any CPU clocking yet. Waiting for Dave's guide to release and do a thorough read over of it.

Dave, like my 2500k build, helped me get 2133 working by changing a few timings.

I am running fully stable with 2133 at 1.5v. XMP on, but my board/bios set one timing, the jFAW too tight so it would freeze and bluescreen with 0x124 code about 20 seconds after hitting the desktop. Changed the jFAW timing from 32 to 40 which is what it should be been set too based on the DIMMs programming, but the board set it to 32.

Once switch to 40, and everything else kept as is, things have worked flawlessly.

I see. I observed some stress test crashes at stock speeds, maybe this was why. I'll change mine to 40 and see if it helps.

 

[TheHunter]

..
I know someone mentioned earlier when SVID is set to Disabled it defaults to 1.7v...that just isn't true...it's a bug/glitch. Because I set to disabled many times before and a couple days since my last CMOS clear and was able to change anywhere between a tested 1.65v and 1.90v with readings in BIOS and AIDA corresponding the change. They both also correspond to the voltage lock.

Just figured I'd share another issue that is causing me instability concerns with the crowd, get some feedback and see what ya'all are doing about it.



Edit: Cleared CMOS, SVID is now adjustable again. Now to see if I can't keep this thing from crashing on every stability test above stock! lol!

Well it is here, it could be motherboard specific and my Deluxe has better control over it.

If i leave it at auto it defaults to either 1.798v or 1.779v, if i set to disabled it defaults to 1.70v or in windows 1.696v.


Like Womper said set it to enabled and put in manually 1.80v or 1.85v.

 

[PolRoger]

Had some luck with 4.3 @ 1.26v, 8-39X Cache @ 1.15v, Agent at + .100v, SVID Disabled / 1.80v, Memory at 2133 XMP, 1.60v (set to 1.58v in BIOS) 11-11-11-28 cr1. It ran Wargame for a couple hours while I watched a movie. Is an hour into a several hour test with Intel's ETU on CPU Test, will do Memory test next. Interesting how it's so stable at 1.26v where it was so spotty before...the System Agent at +.100v is to thank for that at this point, and the SVID being adjusted to 1.80v I believe has a big part to play as well.

Still working on fine tuning...4.3 and 4.4 settings, will also see what 4.2 takes...under stock load I can see upwards of 1.21v iirc...see how far I can get which I'm assuming is 4.2 maybe. It's too bad you can't adjust multi's by the tenth...I'm sure in the next generation or two that will happen... maybe I could hit 43.3X Multi. Of course I suppose if a person wants to mix BCLK and multi's that can be achieved already.

I'm pretty much of the opinion that Haswell runs best (most stable) under ambient cooling conditions with a cpu vcore set in the ~1.175v to ~1.225v range. Which is probably why the ASUS guide recommends starting with vcore @1.200v to gauge/bin the capability of a given sample.

If your sample seems to need ~1.26v for 43x then I think you'll find that your chip will have a overclocking sweet spot at 42x with ~1.200v/1.225v.

My sample needs ~1.259v/1.264v (vid) for 47x and I'm still running into the occasional random BSOD under long term "crunching" loads. If I bump multi up to 48x I can "at best" only achieve a certain level of "benching" stability with a VID in the mid 1.3v range. However my sample seems good (stable) @46x ~1.205v/1.215v and @45x ~1.70v/1.75v range.

I've also been testing 46x with ~101.4/101.5 bclk (4.65GHz/4.66Ghz) using ~1.245v/1.250v VID.

I see. I observed some stress test crashes at stock speeds, maybe this was why. I'll change mine to 40 and see if it helps.

You can download and use MemTweakIt to monitor your ram timings/settings. On my ASUS Z87-Deluxe the 2133 speed (divider/multi) will auto/defaults tFAW to 40.

My current 46x overclock...Using Manual CPU Core/Cache settings (~35+ hrs. Rosetta load):

 

[radrok]

^^^

Seems like a very sweet chip you've there got sir.

May I ask you a Cinebench 11.5 run?

Thanks

 

[PolRoger]

^^^

Seems like a very sweet chip you've there got sir.

May I ask you a Cinebench 11.5 run?

Thanks

Thanks mate...

 

[radrok]

Thank you, that's a very good score.

The IPC on Haswell is impressive, imagine what a higher core count chip would achieve...

 

[Womper]

I'm pretty much of the opinion that Haswell runs best (most stable) under ambient cooling conditions with a cpu vcore set in the ~1.175v to ~1.225v range. Which is probably why the ASUS guide recommends starting with vcore @1.200v to gauge/bin the capability of a given sample.

If your sample seems to need ~1.26v for 43x then I think you'll find that your chip will have a overclocking sweet spot at 42x with ~1.200v/1.225v.

My sample needs ~1.259v/1.264v (vid) for 47x and I'm still running into the occasional random BSOD under long term "crunching" loads. If I bump multi up to 48x I can "at best" only achieve a certain level of "benching" stability with a VID in the mid 1.3v range. However my sample seems good (stable) @46x ~1.205v/1.215v and @45x ~1.70v/1.75v range.

I've also been testing 46x with ~101.4/101.5 bclk (4.65GHz/4.66Ghz) using ~1.245v/1.250v VID.



You can download and use MemTweakIt to monitor your ram timings/settings. On my ASUS Z87-Deluxe the 2133 speed (divider/multi) will auto/defaults tFAW to 40.

My current 46x overclock...Using Manual CPU Core/Cache settings (~35+ hrs. Rosetta load):

Well there you go. The default tFAW for some people is 40 with DDR at 2133. MxPhenom and I have boards that try to use 32.

Are you sticking with Manual mode for voltages? I'm curious if manual mode lets you use lower voltages than Adaptive or Offset.

I just finished a 4 hour IET session at 4.7GHz with 1.275v adaptive vcore, and 1.3v adaptive cache @ 4.5GHz. But 4.8GHz is so tasty, and it's been solid with a manual 1.34v. Think a manual voltage that high would kill my chip quickly? I'm not familiar with how well these 22nm chips can handle what voltage.

 

[DOM]

Have a question for some reason I cant oc for crap on water but in the ss I was gettimg up to 4.9 stable but cant get even past 4 XD

do I need to rma the mb cuz it killed a cpu already on ln2

msi mpower max

 

[Kursah]

You could try to RMA the MB? Hey question? Is your chip Maylay or Costa Rica? I have read that the Costa Rica's are the new ones and are even worse clockers...though I've only read that in a couple of threads...maybe there's something to it. Makes me want to keep my chip and not risk RMA for a worse chip...

4.2 @ 1.21v seems stable so far, had one bsod all night which is decent...none during gaming thus far which is good. Still testing. 4.3 at 1.26v seems mostly stable..going higher with vcore doesn't seem to affect intermittent stability, same with 4.4 @ 1.35v. I'm still thinking I want to make 4.3 where I settle. Should be do-able I hope...but I'm gonna at least make sure I can limp back to 4.2 stable before I go back up.

Also gonna look at my timings for 2133...man wish I could run tighter timings at that speed, but my kit was pretty affordable at time of purchase, fair trade I suppose!

EDIT: My (FOUR ACT WIN Time) tFAW memory setting (section 2 in BIOS) was also defaulted to 32, set to 40 to see if that helps with any of my OC stability issues.

 

[DOM]

Mine is costa from rma I haven't messed with the mem just running stock 2400 9-11-11-28 at T1

might just sell them and get a 3770k

still have my mvg should of sold the haswell rig

 

[PolRoger]

Well there you go. The default tFAW for some people is 40 with DDR at 2133. MxPhenom and I have boards that try to use 32.

That is interesting that the Hero and your Z87-Plus are running/using Auto rules for tFAW that are tighter then my ASUS board? I would have assumed that the same family of chipset/boards would be using similar rules? I've attached some shots of my board's auto defaults for 1600/1866/2133/2400... Primary timings were manually set (4x4GB kit) all others left on auto.

Are you sticking with Manual mode for voltages? I'm curious if manual mode lets you use lower voltages than Adaptive or Offset.

I'm not sure if manual offers an advantage or not?... I would think that the voltage requirements (settings) would end up being rather similar between the three modes. Under my usage scenario this chip primarily runs under 100% load "crunching"... I've also noticed with my current settings that C States are active and my vcore will idle down under C7.

I just finished a 4 hour IET session at 4.7GHz with 1.275v adaptive vcore, and 1.3v adaptive cache @ 4.5GHz. But 4.8GHz is so tasty, and it's been solid with a manual 1.34v. Think a manual voltage that high would kill my chip quickly? I'm not familiar with how well these 22nm chips can handle what voltage.

It is probably too early to say how durable these chips are when running voltages long term at greater then ~1.3v? I've seen some people on the forums who do and don't seem too concerned about the possibility of degradation or other possible damage/death. With IB and Haswell (both 22nm) I personally prefer keeping voltages below 1.3v for daily "24/7" type usage... more like ~1.250/1.275v range. Frankly, 100MHz/200MHz/300MHz isn't really that big a deal under daily real world usage. If you need to bump up your "o.c." short term for a particular reason... Then why not?

 

[TheHunter]

Mine is also 32 at 2133mhz., by 2400, 2666 its ~ 50 and slower speeds, using same 100:133 divider








I didnt test round trip latency (tRTL) 40/41 and (tIOL) 4 yet



@Kursah

Try to disable this if it helps