1. Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Intel Haswell Overclocking Clubhouse.

Discussion in 'techPowerUp! Club Forum' started by cadaveca, Jun 9, 2013.

  1. cadaveca

    cadaveca My name is Dave

    Joined:
    Apr 10, 2006
    Messages:
    14,058 (4.50/day)
    Thanks Received:
    7,223
    Location:
    Edmonton, Alberta
    Lower cache multi to 39

    CPU Cache voltage is dynamic with multi, when default settings are used. Many boards are setting higher-than-default cache speeds and setting voltage to 1.15V this is part of the speed boost some boards see, and is also responsible for slightly elevated temperatures, too. Cache voltage does not seem to boost up like CPU voltage under heavy AVX loading.

    Very few boards have "default Intel settings", so what your BIOS does on your board(in general) is going to depend on quite a few different factors.
     
    Womper says thanks.
  2. Jstn7477

    Jstn7477

    Joined:
    Aug 30, 2009
    Messages:
    3,889 (2.06/day)
    Thanks Received:
    1,568
    Location:
    Sarasota, Florida, USA
    My chip does 4.4GHz @ 1.22v fixed vcore, BUT the cache does 4.3GHz @ 1.23v. I cannot get the cache/ring stable at 4.4GHz even with 1.27v cache voltage. 4.5GHz core needs ~1.28v vcore which is way too much for me.
     
    Crunching for Team TPU More than 25k PPD
  3. Womper

    Joined:
    Jun 10, 2013
    Messages:
    95 (0.19/day)
    Thanks Received:
    45
    Excellent news, I really appreciate your time! With a 1.325v or +0.3v offset, I was worried these AVX loads might jump it into the 1.4v range. Glad that's not the case.

    Now I'm wondering what the Auto setting is doing for Min Cache Ratio. I want to be sure that it's not forcing full cache speed and thus turbo voltage constantly. Does CPU-Z's NB frequency value show the cache slowing down when appropriate?
     
  4. cadaveca

    cadaveca My name is Dave

    Joined:
    Apr 10, 2006
    Messages:
    14,058 (4.50/day)
    Thanks Received:
    7,223
    Location:
    Edmonton, Alberta
    Yes, it does, but as I mentioned, many settings in BIOS seem to affect what happens. Getting proper "stock" settings can be torturous, so I'm rather hestitant to say "YES IT WORKS", when I know for many, it won't.
     
  5. TheHunter

    TheHunter

    Joined:
    Apr 10, 2012
    Messages:
    967 (1.04/day)
    Thanks Received:
    366
    Location:
    Europa
    Last time when I monitored cache frequency (north bridge on pic*) it stayed at fixed multi, min at auto didint change it.

    north bridge.png


    By min 35x; max 42x, it lowered to min 35x in idle state
     
    cadaveca says thanks.
  6. cadaveca

    cadaveca My name is Dave

    Joined:
    Apr 10, 2006
    Messages:
    14,058 (4.50/day)
    Thanks Received:
    7,223
    Location:
    Edmonton, Alberta
    Perfect example of why I was a bit leery of saying cache multi scaling was working. Thanks!
     
  7. Womper

    Joined:
    Jun 10, 2013
    Messages:
    95 (0.19/day)
    Thanks Received:
    45
    Beat me to it, I was going to measure after setting a min value in the BIOS. I'll be interested in the stability ramifications.
     
  8. cadaveca

    cadaveca My name is Dave

    Joined:
    Apr 10, 2006
    Messages:
    14,058 (4.50/day)
    Thanks Received:
    7,223
    Location:
    Edmonton, Alberta
    For boards that offer that option, that works usually, but not all boards do.
     
  9. dumo

    dumo

    Joined:
    Mar 18, 2005
    Messages:
    686 (0.20/day)
    Thanks Received:
    627
    Location:
    NYC
    M6I tests

    Ram @ 3500+

    [​IMG]

    [​IMG]
     
  10. Kursah

    Kursah

    Joined:
    Oct 15, 2006
    Messages:
    7,881 (2.68/day)
    Thanks Received:
    1,714
    Location:
    Missoula, MT, USA
    Well starting to get frustrated by my damn 4770k.

    It seems no matter what now, what was stable a month ago is shit now. I've cleared CMOS, started over. Seems I keep getting the 124 BSOD code which generally means more CPU voltage needed. I'll admit I have a crap clocker. I had been pushing for 4.4GHz (without going past 1.35v) stable on air out of it the last few weeks at an hour here and there when I had time. I get back from a week-long backpacking trip and decided hell with it, I'll go back to my stable 4.3GHz @ 1.23v...hell half the time it'll BSOD going into windows now let alone a stress test. And this is after the overall ambient temps are around 10F cooler than a month ago.

    I just don't get it...but seeing as I bought my chip thru Newegg I think my time's up for any kind of RMA there...I did get the Intel OC warranty...but the chip hasn't failed. It's just a crap clocker. At first it was fun and exciting, now it's just pissing me off. I'm disappointed I can't attain a stable 800MHz OC on a currently for the socket top-end K series chip. This is the worst clocking Intel chip I've seen since my P4. Though it's performance is great, and with that I am happy...part of why I made this build was for overclocking.

    I am just getting frustrated with the inconsistent experience this chip has given me since I got it. I figured maybe it was the MB bios. Updated to most recent. I have tried my RAM down to 1600..didn't change stability. My cache is at 39 and 1.20v set in BIOS atm. Input Voltage is 1.90v at the moment. I have the CPU voltage at 1.25v.

    Sorry for the rant...I was hoping to finally be able to post some more concrete stable results...but my chip just seems to get worse and worse. What was stable at 1.20v needed 1.23v...now 1.25v. And it all comes off of trying to play some Wargame: AirLand Battle before I hit the sack and my last known wPrime/BF3 stable OC with Adaptive Voltage is now unstable.

    I dunno...should I even try to have intel RMA it? Will they without it being dead? I guess I'll have to look into it some more.
     
  11. MxPhenom 216

    MxPhenom 216 Corsair Fanboy

    Joined:
    Aug 31, 2010
    Messages:
    10,066 (6.61/day)
    Thanks Received:
    2,277
    Location:
    Seattle, WA
    I don't know much about Haswell, but at first I was thinking you may be running into a memory issue, but now that you need more voltage for what used to be stable for you, sounds like you may have ran into a case of degradation, or what you thought was stable, never actually was.
     
    Last edited: Jul 30, 2013
    Kursah says thanks.
  12. petedread

    petedread

    Joined:
    May 29, 2011
    Messages:
    274 (0.22/day)
    Thanks Received:
    22
    Location:
    Doncaster
    Mine seems random and inconsistent. But it's probably just me being a noob.
    Anybody using Samsung Green with Haswel?
     
  13. Kursah

    Kursah

    Joined:
    Oct 15, 2006
    Messages:
    7,881 (2.68/day)
    Thanks Received:
    1,714
    Location:
    Missoula, MT, USA
    Ya Mx I was wondering if it's degredation myself. But maybe you're right and it was never really all that stable. But it wasn't this unstable prior and it has just annoyed me at this point. Again thankfully the performance is good, but the OC-ability of this chip has turned out to be much more of a hassle to find than a K chip should be imho. Wish I had money to buy another one and sell this one.

    I will say the Asus board has been pretty awesome thus far, I love the bios and featureset.
     
  14. Womper

    Joined:
    Jun 10, 2013
    Messages:
    95 (0.19/day)
    Thanks Received:
    45
    I'm still running into crashes. It'll go for days without issue and then decide to freeze on me or throw a 0x124 several times in one night. I'm relying on Auto rules for everything besides core/cache voltage, so I'm wondering if those rules aren't quite tuned right. I also wonder if it's the motherboard overheating, but if a 200mm fan isn't good enough, then the motherboard is the problem.

    I've also fought this sandforce incompatibility all along waiting for a BIOS fix, who knows how many bluescreens or freezes it caused. I imaged over to a newer drive, so hopefully that issue is buried. I'm going to run at 4.7GHz core, 4.5Ghz cache with a 1.34v adaptive voltage on both for a while and see if I find any crashes. 4.8GHz core/4.6GHz cache ran for several days on these settings, so running 100MHz less should be stable if it truly just needs more voltage for long term stability.
     
    Kursah says thanks.
  15. petedread

    petedread

    Joined:
    May 29, 2011
    Messages:
    274 (0.22/day)
    Thanks Received:
    22
    Location:
    Doncaster
    @Womper, I bet there isn't another retail chip like yours anywhere on the plannet!
     
  16. petedread

    petedread

    Joined:
    May 29, 2011
    Messages:
    274 (0.22/day)
    Thanks Received:
    22
    Location:
    Doncaster
    I thought about selling mine and buying another but, I recon the odds of getting a better one are terrible. Only slightly better odds than buying a lottery ticket. Well, saying that, it looks like most of the guys on this forum have got good chips. But most of the guys on UK forums haven't got so lucky. So here in the UK the odds look pretty bad.
     
    Kursah says thanks.
  17. Womper

    Joined:
    Jun 10, 2013
    Messages:
    95 (0.19/day)
    Thanks Received:
    45
    From what I've gathered, I think it's a good one aside from long term stability (I've observed even stock frequencies and Auto voltage rules resulting in BSOD in the long term). Keep in mind that adaptive voltage at 1.34v translates to 1.44v during Prime95 and other stress tests. In other words, playing around at 4.7GHz needs a manual voltage of 1.325 to 1.385 (I don't really know what long-term stability actually requires to be honest). At this voltage, heat pretty much prevents me from using lots of these stress tests.

    So my chip is nice in the sense that it can utilize ludicrously high voltages. Safe enough for gaming though.
     
  18. petedread

    petedread

    Joined:
    May 29, 2011
    Messages:
    274 (0.22/day)
    Thanks Received:
    22
    Location:
    Doncaster
    I see. But still, you have a good un. I suppose it gives you a lot of options for 24/7 overclocks. I wonder what volts you need for 4.6ghz? From what I've been reading it seems there is usually a point at which the cpu suddenly needs a big jump in volts. So say 1.325v to 1.385v for 4.7 but maybe only 1.250v for 4.6ghz. 4.6ghz would be a great 24/7 overclock. My cache doesn't like being overclocked, and the cpu doesn't either lol, but you can OC both :)
     
  19. petedread

    petedread

    Joined:
    May 29, 2011
    Messages:
    274 (0.22/day)
    Thanks Received:
    22
    Location:
    Doncaster
    Just to add to the discussion about cache turbo. My cache turbo's to 4ghz if I leave the cache multi at stock settings. If I oc my core to 4.3/4/5 the cache turbos to 4.0ghz. If I oc the cache to 4.3 it turbos to 4.3 and drops back to 3.5 when idle. Or, it did yesterday. I've just had a look and it is now permanently at 4.3 when idle.
    I've uninstalled XTU because it caused some extra inconstancy's.
     
  20. Kursah

    Kursah

    Joined:
    Oct 15, 2006
    Messages:
    7,881 (2.68/day)
    Thanks Received:
    1,714
    Location:
    Missoula, MT, USA
    Ya that's what I'm dealing with for sure. I seem to have found some stability today with 1.27v for 4.3Ghz and 4.0 cache at 1.20v. I have felt my heatsinks and with my cpu cooler venting out the back of the case the MB temps are a few degrees cooler and the VRM heatspreader there is much much cooler. I only tried vertical exhaust for a couple hours before switching back...may try again when I have more time. But I have yet to see my MB sensors go above 35C, and no heatsink has been too hot to touch...this is probably the coolest running MB I've ever owned.

    Maybe it's a BIOS issue? I dunno the recent inconsistencies with my chip has just frustrated me...but I'll get it sorted eventually. At this point I have some stability at 1.27v. It's odd though...I can't even get into windows at 1.23v anymore...and that used to be Prime stable and iirc BF3 stable...but as mentioned earlier it may have not been stable in the first place...but it was more stable then than it is now at those same settings. But degredation could explain that as well...though that seems awfully fast for a new chip imho. Been a loooong time since I've had a bad CPU...but I have had one before so maybe that's the route this one is taking?

    :toast:
     
  21. TheHunter

    TheHunter

    Joined:
    Apr 10, 2012
    Messages:
    967 (1.04/day)
    Thanks Received:
    366
    Location:
    Europa
    I tested 4.6Ghz a bit and this is what I got :)

    cpu multi 46x 100mhz @ 1.22v in bios
    cpu offset 0.002v
    total turbo 1.218v

    cache multi 42x
    fixed voltage 1.13v

    XMP 2133mhz CL9-10-10-27-1T

    input voltage disabled so its fixed at ~ 1.70v
    Cpu TDP 130%
    LLC 6
    VRM optimized

    rest auto
    [​IMG]

    [​IMG] [​IMG] [​IMG] [​IMG] [​IMG] [​IMG]

    small correction: much better score with new wprime32 v2.10
    [​IMG]
     
    Last edited: Jul 31, 2013
    Kursah says thanks.
  22. kracker New Member

    Joined:
    Apr 5, 2013
    Messages:
    2 (0.00/day)
    Thanks Received:
    0
    Location:
    California, USA
    Well, I'm not sure if you use Prime95 for stressing, but the author is optimizing it for Haswell.(FMA3 and AVX2 I believe)

    i5-4670K

    27.9(current) temps on cores
    78 / 74 / 72 / 70

    Future 28.1 temps
    86 / 83 / 80 / 78

    EDIT:...enjoy...
     
  23. Womper

    Joined:
    Jun 10, 2013
    Messages:
    95 (0.19/day)
    Thanks Received:
    45
    Something I like to do is boot into Windows with some "known good" manual voltages and then open Intel Extreme Tuner. Then I crank the core and cache frequencies up and keep voltage low, just to see the absolute floor voltage required to not freeze/bsod.

    I found that at 4.7GHz, I can use 1.28v and get away with 1 or two minutes (max) of Prime95 testing before crashing. For more practical stability, we're talking 1.325 -1.385V. I find it interesting that it needs like 0.1v more than the floor value to be stable...my experience with Nehalem wasn't like this.
     
  24. MxPhenom 216

    MxPhenom 216 Corsair Fanboy

    Joined:
    Aug 31, 2010
    Messages:
    10,066 (6.61/day)
    Thanks Received:
    2,277
    Location:
    Seattle, WA
    Why are you using Prime 95 for stress testing. I thought it didn't officially support Haswell quite yet and the AVX2 instruction?
     
  25. PolRoger

    PolRoger

    Joined:
    Oct 5, 2009
    Messages:
    225 (0.12/day)
    Thanks Received:
    141
    I'm also thinking that it is more likely that your original settings were not completely stable rather then the chip suffering from degradation. It doesn't sound like you were pushing excessive volts if you were mostly at around ~1.25v set in BIOS.

    Like Womper already mentioned about his chip...

    I also have experienced more then my share of "long term" instability with my chip... mostly random BSOD's and sometimes the occasional freeze in my attempts to dial my overclocks. I like to "crunch" Rosetta/WCG so my system pretty much runs 24/7 100% load. I have found Haswell to be more challenging than SB/IB at establishing stable daily "crunching" overclocks but I think I might actually have two that so far seem to be working for me??... One at 46x DDR3-2666C10 and another at 47x DDR3-2133C9.

    I would go back and try starting over again testing at lower speed overclocks...

    Keep your CPU Cache multi on auto (39x) CPU Cache, CPU SA, CPU Digital/Analog I/O voltages all on auto. Run memory at 1600 speed with XMP (defaults/1.6v for your kit/F3-17000CL11D-8GBSR?).

    I would test again for max multi(s) that your chip can run with vcore set in BIOS to ~1.200v/1.225v/1.250v. So maybe something like... 42x@1.2v, 43x@1.225v and 44x@1.25v.



    I don't have the low profile Samsung Green kit but I do have several kits that use similar ic as the Green... (Samsung rev. D). I have some GSkill Trident X sticks that should be using the exact same HYK0 ic as what is used on the Samsung Green. They have run well for me up through 2133C9/2400C9 speeds. My kits can run higher speeds but they are also a higher bin... 2600C10/2666C10. Benching higher speeds with Samsung on Haswell may also require making adjustments to Secondary/Tertiary timings and RTL's in the Memory section of your BIOS.
     
    Kursah and cadaveca say thanks.

Currently Active Users Viewing This Thread: 1 (0 members and 1 guest)

Share This Page