Intel Haswell Overclocking Clubhouse.

[Durvelle27]

^
You will have some trouble with non K model. I think 4.2Ghz is pretty much max.

So far i've manged only 4.173GHz

 

[Batmanjl]

Anyone managed to clock a i7-4770k with a ASrock Z87 extreme 3? if so care to share any info for it?
Anyone have one from batch L319B891 with any luck?

 

[cadaveca]

Dave, I know you're going to test your chips on Z97 boards, but would you please include benchmarks on old Z87 boards now that most manufacturers are releasing Haswell refresh BIOS updates?

Of course. I kept a couple of Z87 boards to review for just this purpose. I'll have to check that out when doing my 2.0 OC Guide, no matter what.

 

[fullinfusion]

A nice little chip this one is.

 

[Womper]

Hey guys, any of you thinking of getting some Devil's Canyon action?

I wonder how they will OC compared to regular Haswell parts...

At $339, 4970K is kinda tempting, my 4770K tops at 4.7, so I was thinking Devil's Canyon should definitely clock north of 5GHz

Both my 2600K and 3930K can hit 5GHz so it was really disapointing to see Haswell top at a lower clock as even at 4.7GHz it already is faster than my previous processors in most tasks due to the higher IPC.

The 4790K engineering sample reviews that have popped up over the past few weeks have not shown anything special. They all basically got to 4.7GHz (stable enough for their benchmarks, who knows if gaming stable or 24/7 stable) with fairly high voltages (high 1.2 to low 1.3). But, the new TIM actually kept temps below throttling even at >1.3v on air, if I recall. Per usual, it's hard to tell if they encountered the +AVX voltage and temps. I was hoping that the thermals were the limit, and maybe with some steppings we'd see 5GHz within reach at voltages in the 1.3 range. It looks like 4.7-4.8 is still pretty much the max you can hope for...but at least with the 4790K it so far seems a guarantee.

The kicker? I was running between +.005-+.010 on Adaptive for 1.151-1.156 and it was NOT stable. Both OCCT's Linpack solo and PSU test would crash/lockup/fail within 15-20 minutes. Temps were higher.

Interesting find, maybe the story has changed with the newer bios. I encountered another random blue screen from Planetside after many months, and bumped up another .01v. That puts me at 1.295v target adaptive voltage, with 0v offset, on both core and cache. Core is at 4.7GHz, cache is at 4.5GHz. All the other voltages are still on Auto.

I also noticed that peak CPU voltage is now hitting the +AVX number, which is in the 1.4v range with my 1.295v adaptive mode. Not sure if it's Planetside or Wildstar doing it. Peak temperature is still about 60C under my Zalman air cooler, so no worries.

 

[TheHunter]

I noticed the latest bios 2003 overvolts cache a bit more then 1707 or 1802

before @ 42x 1.135v would make max total adaptive 1.16v, now its 1.178v or so..


Also older 1802 use to be ok @ 4.7ghz 1.275v, now im back to 1.278v or it could hard freeze in something like BF Hardline.
in both cases,
SVID disabled and manual 1.79v
iVR fault management - disabled
in DIGI+ cpu & ram current 120%, rest auto
cpu system agent 0.055v+ offset for 2400mhz OC ram, otherwise its enough @ ~ 0.020v




But on a side note, Im having some really hard time OC'ing with 125mhz blck, when ever I try ie 125mhz x 38multi (total 4.75ghz) always I see a red light blinking by gpu (@ mobo), the only time I managed to OC via 125mhz is when I first got the mobo and flipped TPU switch to no2.. Anybody know why its doing that? Do I need to OC outside XMP mode?

 

[MxPhenom 216]

I have started messing with overclocking again on my chip after updating the BIOS are starting over. So far just inching it up in increments. at 4.2GHZ at 1.25v and cache at x39 1.15v. Left memory at auto so 1333 to get memory instability out of the equation when I am trying to find my max 24/7 CPU clock. Next ill try 4.2 at the same 1.25v.

I have left everything in the Digi+ power control at auto except for LLC which is at Level 6.

 

[MxPhenom 216]

When increasing the cache multiplier to be within 3-4x of the CPU core multi, is the cache voltage the only voltage that should be taken into account, or can system agent help in that regard as well? I know that if you increase the system agent a bit, you can also lower vcore a tiny bit as well.

I noticed the latest bios 2003 overvolts cache a bit more then 1707 or 1802

before @ 42x 1.135v would make max total adaptive 1.16v, now its 1.178v or so..


Also older 1802 use to be ok @ 4.7ghz 1.275v, now im back to 1.278v or it could hard freeze in something like BF Hardline.
in both cases,
SVID disabled and manual 1.79v
iVR fault management - disabled
in DIGI+ cpu & ram current 120%, rest auto
cpu system agent 0.055v+ offset for 2400mhz OC ram, otherwise its enough @ ~ 0.020v




But on a side note, Im having some really hard time OC'ing with 125mhz blck, when ever I try ie 125mhz x 38multi (total 4.75ghz) always I see a red light blinking by gpu (@ mobo), the only time I managed to OC via 125mhz is when I first got the mobo and flipped TPU switch to no2.. Anybody know why its doing that? Do I need to OC outside XMP mode?

I want your chip lol.

For 4.4 my chip seems to want 1.275v in BIOS and 1.296v @ load with 120% VRM current. Before I had current at auto, but failed prime 2 minutes in, with current at 120% its been going for a bit longer. We will see. I dont want to go over 1.3v. I may settle for 4.3GHZ. It can do it all day at 1.25v.

EDIT: And BSOD. now to try 1.285v 4.4GHZ 120% current.

EDIT: 1.312v load @ 4.4GHZ and 130% current seems to be the sweet spot.

@TheHunter

This guide might help you. Its what I am following and I really like it.

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

It has a whole elaborate section on the BCLK overclocking.

 

[TheHunter]

Yeah higher cpu current ~ 130% can help at higher voltage & cpu frequency.


Thanks will try blck OC one more time, but idk its like there is a bug when it tries to switch to soutbirdge pcie controller, could that be it? This is the only thing I can think of atm..

 

[MxPhenom 216]

Yeah higher cpu current ~ 130% can help at higher voltage & cpu frequency.


Thanks will try blck OC one more time, but idk its like there is a bug when it tries to switch to soutbirdge pcie controller, could that be it? This is the only thing I can think of atm..

You might be onto something, but I am not really sure. I will likely not touch the BCLK. What cache multi you running. I read in this thread from Dave that running it within 3-4 of the CPU multi is optimal. I am at 39x right now with 1.15v for the vcache.

 

[TheHunter]

Im using min 39x and max 42x @ 1.135v adaptive, tested 44x @ 1.17v few times but I didnt see any real difference other then more heat, cinebench15 scored 4points more that's about it.



edit: you can try LLC -auto and 120%, I found out its best to leave all at auto in digi+(since last 3 bios releases), except for cpu & ram current %.

 

[TheHunter]

I did try per core- performance OC again and this time it looks promising

core1 48x
core2 48x
core3 48x
core4 47x

I tested 49x core1,2 and 48x core3,4 not so long ago and saw I need 1.323v for full 4.8ghz load (but its a little toasty for H90), so that being said, I set total 1.295v for 3core 48x and so far it passed all. Tested 3dmark Ice and cloud gate benchmarks, both lightly threaded to see if it will fail when running @ 4.8ghz and it passed all, also played some TF2, MKKE and still no issues (both lightly threaded so it was at 4.8ghz 80% of the time).

Btw now it set full 4.7ghz load @ 1.2727v instead of manual 1.2773v (reading in windows), so I guess it s a win win

Hopefully it sticks this way, Im still using the same settings as above, only cpuv is now at 1.295v.



EDIT:

Asus released another bios and now marked V2003 as beta lol


Z87-DELUXE BIOS 2004 and BIOS updater
Improve System performance

*Note: Must apply the attached BIOS updater tool first before using the new Intel 4th Gen Core processors.



EDIT2:
ok I had to install older IMEI found @ Asus site Version 9.5.15.1730 whole package and now its letting me pass the update tool. Before that I had driver only V10.0. 214 and it wasnt ok..

 

[MxPhenom 216]

what would you guys say is max vcore voltage for these chips with custom water. My max temp has been about ~80c. I feel I might be able to squeeze 4.5.

 

[TheHunter]

depends where you get those 80C,
I try to stay ~ 75C max in cpu stress testing apps., ok something like prime95 or IBT, linx, Aida64 will heat more for sure.
Asus Realbench or x264hd bench or some really cpu/ram demanding game usually shows all instabilities

You can try this per core OC, set first 3 core multi to 45 and last to 44x, this way you will have 4.5ghz by light load and 4.4 full load. But now you will have to raise cpuv a bit, I would say ~1/2 or 1/3 of what you need for 4.5ghz. Mine goes in 0.05v steeps (4.6 1.23 > 4.7 1.278 > 4.8 1.325v), so ~0.025v should be enough.


Speaking of this new bios, IME FW is still the same 9.0.30.1482 as in Feb 2014, but now after I updated to 2004 i cant use this per core OC anymore, well not at the same settings and core3 48, core4 47x and same 1.295v..

I made sure yesterday that it ran fine by light load, but today after I updated to 2004 it kept crashing and wmea bsod'ing in Re5 bench.. I reverted to old 2003 and now its also not stable anymore, needs more cpuv then usual , kinda like in old days.. although no Im back to 2004 again lol D:

EDIT: it was cpu power current fault, 120% wasnt enough anymore now at 130% and Im able to run 4.7ghz @ 1.275v.


Cadaveca do you know what might have changed in 2004 with this "Improve System Performance"?

 

[MxPhenom 216]

I was getting the ~80c during the 2nd test in Prime95, after then it drops and sits at around 65c.

 

[MxPhenom 216]

Ill probably leave it at 4.4GHZ.

I want to dial in memory at 2400. What timings are you using for your memory at 2400?

EDIT: I can run games and anything all day right now at 4.4ghz @ 1.312v. But this is with memory at auto so 1333mhz, to isolate any instability from memory. So if I do get any instability I know its CPU and not memory/IMC.

 

[TheHunter]

Yeah use 4.4ghz then, its still ok


Enable xmp 2133mhz and see how it goes, if it fails try to raise vccsa (cpu system agent) a little, 0.010 or up to 0.030v,

if you want to OC ram then you need to use DIGI+ ram current power @ 120%,
by in ram timings try CL10-12-12-31-1T & tRFC 214, keep rest at auto @ 1.60-1.65v << this ram OC will need more vccsa, 0.040 - 0.055v+ offset is a good start.

I need 0.050v @ 4.7 & 2400mhz, at stock 2133mhz 0.015v+ if i want lower cpuv and stable fps without any fps drops - ResidentEvil5 (jobthread=8)- fixed benchmark by ~ 45-55%.

 

[cadaveca]

I want to dial in memory at 2400. What timings are you using for your memory at 2400?

Check my BIOS screenshots a page or two back for timings. Those were done with my Dominator Platinums.

 

[MxPhenom 216]

Check my BIOS screenshots a page or two back for timings. Those were done with my Dominator Platinums.

Damn those are some tight timings for 2400.

I see for 2400, you left System agent, and the i/o's at auto. Is that best to do unless you have instability? Increase them a bit. I remember reading, (I think you mentioned it) to keep system agent just below the i/o voltage values.

 

[MxPhenom 216]

Went with VCCSA at +0.045 and so far so good. Havent done any real test, but so far windows is functioning fine.

Okay, ran Prime95 blend and VCCSA increased the temps a shit ton. temps maxed at 100c, and when I saw that, I stopped the test.

I think my chip has horrible IHS contact. I don't think it should be getting that hot with my water cooling.

Dropped it down to XMP Profile 2133. But dropped the Command Rate timing from 2 to 1.

EDIT: Still sends the temperatures sky rocketting.

 

[cadaveca]

Damn those are some tight timings for 2400.

I see for 2400, you left System agent, and the i/o's at auto. Is that best to do unless you have instability? Increase them a bit. I remember reading, (I think you mentioned it) to keep system agent just below the i/o voltage values.

Yeah, on this board, that's the stock values, so no need to adjust.

And those timings aren't that tight, that's the normal timings for these sticks at 2400. Yours should do it fine too.

 

[MxPhenom 216]

Yeah, on this board, that's the stock values, so no need to adjust.

And those timings aren't that tight, that's the normal timings for these sticks at 2400. Yours should do it fine too.

Seems, for my chip atleast, memory overclock has a huge impact on the CPU temps.

I had to drop my cpu to 4.2 @ 1.25v to be able to do my memories stock 2133 to keep temps below 90c. Do 4.4GHZ at 1.312v and memory at 2400 VCCSA +0.045 and temps go to 100c and throttle.

Temps never got higher then 85c at 4.4GHZ 1.312v with memory at auto (1333)

 

[cadaveca]

vINPUT is likely on AUTO, methinks. try setting 1.760 like I have at first, with just the CPU OC, if CPU OC then fails, then increase. Likewise, when you find stable, maybe try increase a bit, then decrease vCPU

 

[MxPhenom 216]

vINPUT is likely on AUTO, methinks. try setting 1.760 like I have at first, with just the CPU OC, if CPU OC then fails, then increase. Likewise, when you find stable, maybe try increase a bit, then decrease vCPU

vinput is on auto yes. Can that voltage option push temps pretty good?

I'll play around with it a bit more next week. I need to get ready for class in the morning.

 

[TheHunter]

Well dont test Prime95 then, that will heat no matter what.

Also yea I heard higher ram can heat cpu more, that ram current 120% will also heat extra 3-5C vs 100%, but I found out I need at least 110% to be "stable" at 2400mhz or it can hard freeze or reset pc.

Cpu current 130% will heat more too vs 120% (3-5C), but sometimes it just needs 130%...


Vccsa also a little but not much, 2-3C maybe.


I set SVID to disabled and then manually entered in 1.79V., 1.76v was to little for 4.7ghz @ 1.27xV, also iVR fault management to disabled.