Adjusting individual cpu settings and stability testing

Author
sfield
Superclocked Member
  • Total Posts : 143
  • Reward points : 0
  • Joined: 2009/07/19 17:57:10
  • Status: offline
  • Ribbons : 3
2010/10/27 23:47:21 (permalink)
I wanted to adjust the CPU voltage settings to get these as low as possible for each CPU, while maintaining full stability.
 
I'm running at 23x185 (4.255ghz), 19x uncore, DDR3-1850.  I started out at vcore=1.3625v (no vdroop), 1.325v VTT on both processors.  I've since been able to reduce the voltage on one processor, and the VTT on both processors while maintaing stability.
 
CPU0 VCore=1.3625v, VTT=1.30v
CPU1 VCore=1.3375v, VTT=1.30v
 
Beyond the obvious benefits of reducing voltages -- at new settings:
CPU0, delta between idle wall power draw and IBT stress (of only CPU0) is 225W. 
CPU1, delta between idle wall power draw and IBT stress (of only CPU1) is 211W.
 
For my stability testing, I chose to use IntelBurnTest, starting at Standard, and then working up to very high, extreme.  However, the approach documented below should work well for most stability testing tools -- assuming the tests don't mess with the affinity.  Folding_at_home should be possible to test in this manner as well (eg: in conjunction with -smp 12 / -smp 8).
 
Step 1
Fire up Eleet, since you'll want to use this for lowering/tweaking voltage settings -- one processor at a time.
 
 
Step 2
Fire up one copy of Intel Burntest and select the number of threads corresponding to one physical processor (typically 12 or 8).  Move this copy to the left of the eleet window.  Now, fire up windows task manager, select the processes tab, and right click on the IntelBurnTest image, Set Affinity, uncheck All Processors, and then select every processor on Numa Node 0.  That should be half of the total visible processors/threads.  If you have disabled NUMA, just select the first half of the processors.

 
 
Step 3
Fire up another copy of Intel Burntest and select the number of threads corresponding to one physical processor (typically 12 or 8).  Move this copy to the right of the eleet window.  Now, fire up windows task manager, select the processes tab, and right click on the (new) IntelBurnTest image, Set Affinity, uncheck All Processors, and then select every processor on Numa Node 1.  That should be half of the total visible processors/threads.  If you have disabled NUMA, just select the last half of the processors.

 
 
You should now have something that looks like the below.

 
 
Start by running the copy of intel burntest on the left of the eleet window, and then try lowering the CPU1 Vcore while re-testing and maintaining stability.  Once you are happy with those results, move onto running the copy of intel burntest on the right side of the eleet window, and then try lowering the CPU0 Vcore, etc.  After both are verified as being stability individually, run both instances of Intel Burntest, and you can also move onto moving an instance with 32 threads (given there is no 24 thread option).
 
Here's an overlay of some of the relevant voltage, temperature, and NUMA labels, since there is some inconsistency in what constitutes processor 0 and processor 1 depending on where you look.

 
I'm interested to hear results others see.  I was pretty surprised by the difference in vcore between the two processors.  I also measured the voltage using the pads on the motherboard:
vcore0 1.36v vtt0 1.336v -- measured.   vcore0 set to 1.3625v, vtt=1.30v in bios.
vcore1 1.332v vtt1 1.315v -- measured.   vcore1 set to 1.3375v, vtt=1.30v in bios.
 
CPU0 VCore=1.3625v, VTT=1.30v
CPU1 VCore=1.3375v, VTT=1.30v 
  
  
 
post edited by safield - 2010/10/28 01:42:00

Attached Image(s)

#1

8 Replies Related Threads

    may i be worthy
    ACX Member
    • Total Posts : 263
    • Reward points : 0
    • Joined: 2010/07/27 00:20:41
    • Status: offline
    • Ribbons : 2
    Re:Adjusting individual cpu settings and stability testing 2010/10/28 00:26:07 (permalink)
    Again, great stuff, I was going to try this but never got around to it.
     
    Glad you also found that CPU0 as marked in documentation and on the motherboard is actually CPU1 in all software reported utilties. What a PITA.
     
    But what is exciting is that you have been able to reduce the voltage on the hottest CPU - win win!
     
    Just to be sure, when you measured the vcore - what was the input in the bios at the time - your starting point of 1.3625v each? or your final result of 1.3625v and 1.3375v?
     


    SR-2 #3 -Folding/render: Dual Hexacore X5680: @4.301GHz, 12GB @ 2:10 DDR 1850.
    SR-2 #2 -Folding/render: Dual Hexacore X5660 @4.301GHz, 12GB @ 2:10 DDR 1850. | P2686 : 162,850 ppd  
    SR-2 #1 My main work rig. Dual Hexacore X5650 @4.202GHz, 24GB @ 2:8, All aircooled: Noctua DH-14 

    #2
    sfield
    Superclocked Member
    • Total Posts : 143
    • Reward points : 0
    • Joined: 2009/07/19 17:57:10
    • Status: offline
    • Ribbons : 3
    Re:Adjusting individual cpu settings and stability testing 2010/10/28 02:25:41 (permalink)
    I measured with the new/final results.  The only value that was a little suprising was the vtt0 value being a fair amount higher than set.
     
    may i be worthy

    Again, great stuff, I was going to try this but never got around to it.

    Glad you also found that CPU0 as marked in documentation and on the motherboard is actually CPU1 in all software reported utilties. What a PITA.

    But what is exciting is that you have been able to reduce the voltage on the hottest CPU - win win!

    Just to be sure, when you measured the vcore - what was the input in the bios at the time - your starting point of 1.3625v each? or your final result of 1.3625v and 1.3375v?



    #3
    KONAn
    ACX Member
    • Total Posts : 338
    • Reward points : 0
    • Joined: 2009/04/05 11:44:46
    • Location: Gdansk, Poland
    • Status: offline
    • Ribbons : 3
    Re:Adjusting individual cpu settings and stability testing 2010/10/28 14:12:15 (permalink)
    safield

    I measured with the new/final results.  The only value that was a little suprising was the vtt0 value being a fair amount higher than set.
     


    ...and that is what concern me the most...when you set 1.4Vtt in bios it may set 1.43V...and this is way too much for 32nm cpu's...

    EVGA SR-2
    Dual Xeon E5620@3.8 GHz (19*200BCLK) Vcore: 1.33V, Vtt: 1.35V, IOH: 1.4V 
    48GB (12x4GB) Kingston KVR1333D3N9K2/8G @1200MHZ 1.5V@1.6V 7-7-7-20-2T 
    Corsair AX1200, EVGA GTX 285, LSi 9260 8i, Obsidian 700D
    CPU/GPU/MB watercooled
    .
    #4
    jmtyra
    New Member
    • Total Posts : 12
    • Reward points : 0
    • Joined: 2008/08/17 06:39:44
    • Status: offline
    • Ribbons : 0
    Re:Adjusting individual cpu settings and stability testing 2010/10/29 11:00:54 (permalink)
    Excellent info safield -- Will have to give this a try myself.

    Reference Builds: EVGA SR-2 & ASUS Z8NA-D6C -- My Heatware

    #5
    may i be worthy
    ACX Member
    • Total Posts : 263
    • Reward points : 0
    • Joined: 2010/07/27 00:20:41
    • Status: offline
    • Ribbons : 2
    Re:Adjusting individual cpu settings and stability testing 2010/10/31 20:16:18 (permalink)
    Thanks again Safield, as posted on [H]:
     
    Last  night I dropped voltage on CPU-1 (socket labeled) from 1.400v to 1.375v vcore - reducing temps on my hottest cpu by 5+ degrees.
     
    So the end result is that the temps between CPUs are far closer, but the reported voltages are further apart!
     
    CPU-0 socket labeled = 1.400v vcore = reported 1.40v = average 58c folding ("CPU-1")
    CPU-1 socket labeled = 1.375v vcore = reported 1.35v = average 63c folding ("CPU-0")
     
    So thanks for the heads up, and helping untangle a very confusing board!
     
     


    SR-2 #3 -Folding/render: Dual Hexacore X5680: @4.301GHz, 12GB @ 2:10 DDR 1850.
    SR-2 #2 -Folding/render: Dual Hexacore X5660 @4.301GHz, 12GB @ 2:10 DDR 1850. | P2686 : 162,850 ppd  
    SR-2 #1 My main work rig. Dual Hexacore X5650 @4.202GHz, 24GB @ 2:8, All aircooled: Noctua DH-14 

    #6
    texinga
    CLASSIFIED ULTRA Member
    • Total Posts : 5121
    • Reward points : 0
    • Joined: 2010/07/03 14:30:32
    • Status: offline
    • Ribbons : 20
    Re:Adjusting individual cpu settings and stability testing 2011/01/23 07:23:49 (permalink)
    I wanted to use your process, but cant set the "affinity" from the IBT process in task manager (Win7).  I can see the list with the "affinity" selection when I right-click the process.  But when I select "affinity" I get a pop-up window that says "access is denied".  Is there something I need to do somewhere to get around that problem?  I'm running IBT version 2.50.
     
    Rick



    #7
    sfield
    Superclocked Member
    • Total Posts : 143
    • Reward points : 0
    • Joined: 2009/07/19 17:57:10
    • Status: offline
    • Ribbons : 3
    Re:Adjusting individual cpu settings and stability testing 2011/01/23 11:04:19 (permalink)
    When you ran IBT, did you run it elevated as administrator?  You don't need to run it as administrator.  If you want to change the affinity of any process in task manager, click "Show processes from all users" in the processes tab, then right click->set affinity.  This runs task manager as administrator and hence you can change the affinity of processes not running in the standard user context.
     
    Let us know your tuning results.
     
    texinga

    I wanted to use your process, but cant set the "affinity" from the IBT process in task manager (Win7).  I can see the list with the "affinity" selection when I right-click the process.  But when I select "affinity" I get a pop-up window that says "access is denied".  Is there something I need to do somewhere to get around that problem?  I'm running IBT version 2.50.

    Rick


    #8
    texinga
    CLASSIFIED ULTRA Member
    • Total Posts : 5121
    • Reward points : 0
    • Joined: 2010/07/03 14:30:32
    • Status: offline
    • Ribbons : 20
    Re:Adjusting individual cpu settings and stability testing 2011/01/23 12:17:44 (permalink)
    I was not running IBT as Admin when I had the issue with setting affinity, but I also tried running IBT as Admin and had the same issue.  You gave me something new to try (clicking on "Show processes from all users"), so I gave that a try, then tried to set the affinity and it works now.  I don't know what that did, but it worked.  When I go back and try the 3.8G OC again, I'll use your process to see how each CPU performs and also to see if I can narrow down that horrible squeal if it happens again.
     
    Thanks for the help,
     
    Rick



    #9
    Jump to:
  • Back to Mobile