LN2 Adventure - QX6850 + Gigabyte P35-DQ6

Discussion in 'Extreme and Water Cooling' started by T_M, Oct 9, 2007.

  1. T_M

    T_M Member

    Joined:
    Jun 28, 2004
    Messages:
    2,164
    Location:
    Abu Dhabi (ex-Adelaide)
    T_M - Team.AU member

    Over the weekend i finally got around to some LN2 benching here in Singapore.
    My last adventure with LN2 was in Gigabyte's Beijing OC Comp on E6550, and i was keen to give it a go now on proper hardware.
    Still sticking with the P35-DQ6 which i had fully modded for the Beijing competition.
    Here's my setup:

    CPU: QX6850 L724A 'G0'
    Motherboard: Gigabyte P35-DQ6 F5 BIOS thanks Gigabyte
    Cooling: Mousepot Rev1

    Video Card: Riva TNT2 PCI
    Memory: Crucial Value RAM (D9GKX)
    PSU: Silverstone 1000W
    HDD: 500Gb WD AAKS SATAII

    vCore: 1.83v Real (1.84V load) Board will not boot with higher, but can idle with higher
    vDimm: 2.5V Real
    all other voltages at max BIOS settings.
    CPU definately needs more volts than the board can provide.

    The day was all about testing my newest family member the QX6850.
    It had previously shown promise on DICE, but it was clear that DICE couldn't handle the heat load to maintain good benching clocks.

    Here's a video of my first pull down for the day.
    Took approx 10mins to reach -120C.



    Following the pull down, i set about testing for max CPU benchable clocks. Using 32M wPrime on all 4 cores, i bumped clocks starting from 4.8GHz right up thru 5GHz.
    It was clear i was going to have a good afternoon :D

    Next up i set some 32M wPrime and 1024M wPrime times at 5GHz just to get some 'scores on the board' - as any seasoned bencher knows its worth getting at least some scores recorded before something might go wrong. fortunetaly for me, nothign went wrong that day :)

    Next up was checking max suicide clocks at 4 cores. And wow, was i ever suprised. 5400MHz without too much trouble at all.
    And with that under my belt i moved back onto wPrime, one of my favourite benches. Slowly bumping clocks, i edged above 5GHz. That when i encountered what i call 'transition'.

    Transition
    While this is not new to other LN2 benchers, it is to me so i'd like to write up my experiences with it.
    Until now, my LN2 benching had been relying totally on idle temps. Between 32M wPrime runs i had to allows rest time for the system to recover in temps, and during load on 32M wPrime i would lose 10C base temp, no matter if i topped up the pot or not.

    Mid way thru the session, i decided it was time to head to the toilet, so i topped up the pot with an extra large dose of LN2 and made my way down the passage. From the toilet i could still hear the system and bubbling, but i noticed it get very very loud all of a sudden - so much so that i had to cut my toilet break short to check wtf was going wrong. I returned to find the pot looking completely full to the brim, then just as suddenly it dropped in level again to about 1/5th full (where it should have been).
    Temps on the pot had dropped crazy low, i think it was -160C when i checked and dropping fast. It peaked at -19xC and sat there for a long time even with the pot empty.

    Once the pot had finally resumed to a workable temperature, i found that the pot behaved like a totally different beast, holding load was no longer a problem because i could easily move temperature up or down with little pours, and the base no longer lost temp under simple loads like 32M wPrime. Following the session, i spoke to Kinc about my experience and he confirmed it is normal to take a pot down thru this "transition" at -145C or -150C just to change the behaviour of the copper/pot, then once you bring it up again it'll handle totally different!

    So, being the budding youtuber i captured another transition on camera the next day :)
    Be sure to listen to the sound (excuse the background TV) at the tube approaches -152C, then watch the violent change and subsequent "peaceful" rapid drop in temps. Notice also that the bubling changes, before the transition it is larger violent bubbles, after transition it is like a very fine fizz.



    After all of that, i spent a further 3hrs or so benching my QX6850 on this seemingly strong board from Gigabyte.
    Cold bug on this chip appears to be around -130C, i can sometimes go lower and it'll survive, othertimes instant hang. Ive even had base temp at -154C while mid bench and it survived.

    My interest was mainly in 4 core benching, since this chip is a powerhouse intended for some 3D'06 and wPrime damage. I wasnt going for fast scores so my RAM wasn't optimised at all, nor any tweaks.

    I restarted with my "new" pot down at 4.9GHz, wPrime 32M benching my way right up to 5050MHz, keeping the temps between -120C and -130C with small pours prior to each run. If i found that the temp was going to drop below 130C by the idle after run, i'd quickly hit SuperPI 32M to load up the processor while i did online entry of the score.

    wPrime 32M
    http://www.hwbot.org/result.do?resultId=654025

    Then at 5050MHz i moved onto a 1024M run, the way the system felt was like it wasn't the max my chip could do, but i was getting itchy feet to get some scores on the board again.

    Here's a video of my pouring requirements during a heavy loaded 1024M run. As you can see the "new" pot could take whatever the chip could dish out :devil:



    wPrime 1024M
    http://www.hwbot.org/result.do?resultId=654083

    Next up, i moved onto CPU-Z validation for max suicide with 4 cores.
    This is where things start to get really cool hehe

    [​IMG]

    Looking at the difference between this massive clock and my wPrime max clocks, it was clear to me that the CPU just couldnt get enough volts to power thru 4 core benches.

    Now for some SuperPI loving (not my favourite bench these days).
    Totally unoptimised runs, all 4 cores and no affinity:

    [​IMG]

    [​IMG]

    And finally i'll finish of with some Hexus PiFast for the purests:

    [​IMG]

    Here's some photos from the day:

    [​IMG]

    [​IMG]

    Stay tuned for big things with this CPU :woot:
    Thanks for listening.

    Also, i'd like to thank my fellow Team.AU teammates, Gigabyte, Kinc and Hicookie for all their support and tips and tricks over the last few weeks. And Jason (MickeyMouse) for his great pot still going strong.
     
    Last edited: Oct 11, 2007
  2. dirtyd

    dirtyd Member

    Joined:
    Jan 4, 2006
    Messages:
    3,895
    Location:
    Melbs
    Nice work mate. That transition thing was interesting. Sure is a noisy bastard before it hits that point! When are you gonna do some fully optimised runs with everything maxed?
     
  3. Smokin Whale

    Smokin Whale Member

    Joined:
    Nov 29, 2006
    Messages:
    5,175
    Location:
    Pacific Ocean off SC
    :tongue:

    5ghz quad... holy jesus.

    Slap a 8800 ultra sli setup in there and OC that and break some records :p
     
  4. K3nii

    K3nii (Banned or Deleted)

    Joined:
    Jun 4, 2007
    Messages:
    562
    Location:
    Your FACE!
    excellent work u deserve a medal buddy :)
     
  5. zaim

    zaim Member

    Joined:
    Oct 19, 2004
    Messages:
    491
    Location:
    Sunshine, MELBOURNE
    I applaud anyone that can do an <10s superpi 1mb run
     
  6. wwwww

    wwwww Member

    Joined:
    Aug 22, 2005
    Messages:
    4,987
    Location:
    Melbourne

    :) It's my favourite too! :p
     
  7. eva2000

    eva2000 DDR1/DDR2/DDR3 Addict

    Joined:
    Jun 27, 2001
    Messages:
    21,902
    Location:
    Brisbane, Australia
    congrats Tim :D

    when ya gonna make v1.55 wprime which is as fast as v1.51 ? :D
     
  8. n3wbi379

    n3wbi379 Member

    Joined:
    May 15, 2006
    Messages:
    4,600
    Location:
    Melb
    Well done T_M as usaul GREAT STUFF :thumbup:

    whens the qx9650 coming :D
     
  9. Rabe

    Rabe Member

    Joined:
    Mar 29, 2005
    Messages:
    1,389
    Location:
    Sydney
    Wow, really entertaining watch :D Good work.
     
  10. OP
    OP
    T_M

    T_M Member

    Joined:
    Jun 28, 2004
    Messages:
    2,164
    Location:
    Abu Dhabi (ex-Adelaide)
    wwwww, your wPrime latest versions need some work.
    since 1.51 they've all been consistently slower, and some versions (e.g. 1.51) require later versions to be installed in order to work (1.53 installed so that 1.51 will run)
     
  11. wwwww

    wwwww Member

    Joined:
    Aug 22, 2005
    Messages:
    4,987
    Location:
    Melbourne
    1.51 had a bug - where it errors during load and leaves you at the console - you can get passed this by typing
    install mswinsck.ocx
    then
    gui

    You only need to do this once.


    @eva
    v1.53 (or was it 52) and up has a little more cheat detection in it, which slowed the calculation down. It's easy to revert back, but I'd lose this cheat detection.
     
  12. youngpro

    youngpro Member

    Joined:
    Jun 18, 2006
    Messages:
    1,451
    nah its good how it is, doesnt matter if its slower as everyone will use the same version, cheat detection is the main concern!
     
  13. wwwww

    wwwww Member

    Joined:
    Aug 22, 2005
    Messages:
    4,987
    Location:
    Melbourne
    (y)
    The checksum will fail if you use an older version (can use 1.53/1.54 but not any of the faster ones).

    Also I've read on hwbot forums that I'm going to make 1.55 send the version number to hwbot.
     
  14. Scotty1991

    Scotty1991 (Banned or Deleted)

    Joined:
    Apr 27, 2007
    Messages:
    1,527
    Location:
    Sydney
    Awesome read, good videos, thanks :thumbup: :leet:
     
  15. eva2000

    eva2000 DDR1/DDR2/DDR3 Addict

    Joined:
    Jun 27, 2001
    Messages:
    21,902
    Location:
    Brisbane, Australia
    Yeah i figured as much.. but i do hope you can figure out a way to have both cheat detection and faster or on par with v1.51 :D
     
  16. Dezza Bot

    Dezza Bot Member

    Joined:
    Aug 16, 2004
    Messages:
    4,609
    Location:
    Central Coast, NSW
    Am I the only one that read the 'QX65850' and thought, omg new super fast Intel cpu omg etc? Wish my g0 qx6850 would do more than 300mhz without having to send the vcore through the roof.

    Nice results there!
     
  17. ojk007

    ojk007 Member

    Joined:
    Oct 28, 2006
    Messages:
    1,273
    just thought the exact same thing
     
  18. tuchan

    tuchan Member

    Joined:
    Aug 23, 2006
    Messages:
    615
    Location:
    Adelaide
    Nice work indeed Tim. Gotta love the vids for validation!;) :thumbup:
     
  19. hotsuma

    hotsuma Member

    Joined:
    Apr 4, 2007
    Messages:
    1,076
    Location:
    Melb, CBD
    more vids tim keep em coming hehe
     

Share This Page