Core i7/x58 - Dealing with Elpida 50mm Hyper ICs and overclocking

01:13 PM

I've been testing 2x sets of 6GB Corsair Dominator GT DDR3-2000Mhz 7-8-7-20 rated triple channel kits on DFI UT X58-T3EH8. First set had interesting and weird stability issue with single super pi 32M having a lower max overclocking frequency than running 4x or 8x super pi 32m pi via HyperPi or 3dmark benching. Single Super Pi 32M would have lower frequency max than more stressful apps and either freeze or reboot with Elpida 50mm Hyper ICs. Or single Super Pi 32M would be inconsistent, passing 32M pi one instance and next run = reboot/freeze.

A few in the know folks also reported such issues at http://www.ocxtreme.org/forumenus/showthread.php?t=4542. I mentioned this to Gary at Anandtech.com and he's been looking at this with both a Corsair provided Dominator GT kit as well as a retail purchased set. Gary reported some interesting findings regarding Round Trip Latency at http://www.xtremesystems.org/forums/...&postcount=446

Originally Posted by bingo13

Beware of RTL (Round Trip Latency) settings up high now also, they might need finite adjustment based on the board's BIOS settings, IMC quality, memory type/settings, and IOH setup . Once you get to this level, the amount of interaction required to get a stable system is rather painful. I recreated your problem on the DFI board last Friday, switched over to the Gigabyte Extreme and RE II now for clocking this memory up.

Jody/3oh6 who also experienced same single super pi 32m issue also messed with DQ VREF on EVGA as well http://www.ocxtreme.org/forumenus/sh...4&postcount=67 and helped somewhat.[/list]

Originally Posted by 3oh6

i have finally gotten some time to sit down with my Elpida kit on the EVGA X58-SLI (previously dead stuck at 960MHz 7-8-7) with the new BIOS SZ21.

again, i found myself stuck at 960MHz'ish so i started testing each stick individually.
one stick first slot - pass single 32M @ 970 7-8-7-20-1T
two stick f&s slot - pass single 32M @ 970 7-8-7-20-1T
three stick all slots - freeze/hange/reboot/etc... single 32M @970 7-8-7-20-1T

so i simply started empirically testing settings because nothing had worked, low and behold, DIMM DQ Vref seemed to be the secret. everything
else the same, VTT @ 1.44v, vDIMM @ 1.68v under load, i just cracked out a single 32M at 993MHz 7-8-7-20-1T and have ran maybe 15 single
32Ms on the way up from 970 without a single freeze/reboot/restart/bluescreen. i have had a couple errors when CPU frequency got too high
with too little vCORE, but that is what i was hoping for...actual errors, not freezes. just got another actual error on 3rd loop. bumped VTT and just
kicked out a single 32M @ 1000MHz 7-8-7-20-1T.

now i know i just ended my run and ruined the rest of my night posting about it - because that always seems to happen when i post findings and i have no idea how this might relate to other boards...but going from 0 to +100mV DIMM DQ Vref has gotten me over a major hump with my
Dominator-GT 1866s and this board. it has also completely eliminated the huge vDIMM droop i was getting on this board before. HTH someone else out there, i was pulling my short white hairs out with this board and high mem clocks.

CapFTP also reported tweaking ADDR/CMD Bus VREF values helped.

Originally Posted by CapFTP

thnx Jody i like to see someone else got some results..even without speaking each other......i also played yesterday evening after discussing with Hiwa. I finally had my Avantium Elpida pi32 stable on DFI UT x58 + bios 227.
Killer point for me were this settings:

DIMM 1/2 DQ/DQSTB Bus VREF: -00.0%
DIMM 3/4 DQ/DQSTB Bus VREF: -00.0%
DIMM 5/6 DQ/DQSTB Bus VREF: -00.0%
ADDR/CMD VREF Control: Disabled
x DIMM 1/2 ADDR/CMD Bus VREF: 110
x DIMM 3/4 ADDR/CMD Bus VREF: 110
x DIMM 5/6 ADDR/CMD Bus VREF: 110

I came to that though because on Asus r2e i played a bit with vref too.
on DFI i noticed some improvement running this way:

ADDR/CMD VREF Control: Enabled
x DIMM 1/2 ADDR/CMD Bus VREF: 90
x DIMM 3/4 ADDR/CMD Bus VREF: 90
x DIMM 5/6 ADDR/CMD Bus VREF: 90

This helped me a lot. Anyway I had to give Dram voltage = 1.755 from bios (about 1.7 read from SG) which is a lot higher than what I gave on R2e. But this way i reached:

http://news.tecnocomputer.it/stuff/F...2_2000_777.JPG

Now to my own experience, only the first 6GB Corsair Dominator GT kit of my experienced this weird single super pi 32M issue when past 940-965Mhz 7-8-7-20, the second Dominator GT kit has no problems with single super pi 32M tested upto 1000Mhz 7-8-7-20 1T at 1.65v.

Now, testing 2nd kit >1000Mhz 7-8-7-20 1T, I can confirm on DFI UT X58-T3EH8 with 2/27 beta bios that ADDR/CMD Bus VREF bios option did help me get single super pi 32m stable with ADDR/CMD Bus VREF = 95 when it wasn't stable (reboots) with default ADDR/CMD Bus VREF = Disabled @DDR3-2140Mhz 7-8-7-20 1T with 1.61v CPU VTT and 1.875v VDIMM.

Gary (bingo13) mentioned Round Trip Latency (RTL) tweaking too, and I noticed that RTL value in CPU Tweaker loosened up when I changed ADDR/CMD Bus VREF from disabled to enabled with 95 value. RTL changed from 57 to 58 after enabling ADDR/CMD Bus VREF with value of 95.

ADDR/CMD Bus VREF disabled where RTL = 57

ADDR/CMD Bus VREF Enabled = 95 where RTL = 58

So I think we're onto something which will help overclock Elpida 50mm Hyper ICs even further on X58 platform. Now to see what motherboard bios engineers do for future bioses ?

Feel free to share your comments, questions or have more insight of your own Elpida 50mm Hyper ICs experiences on X58 platform ?

Update: May 4th, 2009

Does seem that board bios can also play apart in Elpida IC handling. DFI new 4/28 beta rocks for Elpida 50mm Hyper ICs > 4/28 Beta Bios results - DFI UT X58-T3EH8

4//28 looking sweet indeed 8x 32M Super Pi instance via HyperPi @DDR3-2080Mhz 7-8-7-20 1T at 1.725v bios set vdimm = 1.730v real DMM vdimm.

18-03-2009, 12:03 PM #2 (**permalink**)

Best I can do for now with new found ADDR/CMD Bus VREF tweaking (set at 105) @DDR3-2209Mhz 7-8-7-20 2T at 1.92v vdimm bios set (1.925v real DMM).

@DDR3-2209Mhz 7-8-7-20 2T at 1.92v vdimm bios set (1.925v real DMM)

Validation is at DDR3-2209Mhz just cpuz clock momentarily jumped to DDR3-2212Mhz

@DDR3-2286Mhz 7-8-7-20 1T at 1.875v vdimm bios set (~1.880v real DMM)

Still not close to my best I did when i first got my GT set with DFI UT X58-T3EH8 with 1/23 beta bios.

Validation

18-03-2009, 03:43 PM #3 (**permalink**)

Thanks for the report and info George , i need to start testing my GTs again for comparison.

18-03-2009, 04:19 PM #4 (**permalink**)

You're welcome, I have a 6GB Kingston DDR3-2000Mhz 8-8-8-24 Elpida 50mm Hyper kit as well which exhibits same single super pi 32m issue once past 980mhz 7-8-7-20 1T (then again it's not rated for 7-8-7-20), so will see how the ADDR/CMD Bus VREF tweaking helps there too.

19-03-2009, 10:55 AM #5 (**permalink**)

great result eva!!! I think your ram with chip elpida are the best for intel core i7

19-03-2009, 11:00 AM #6 (**permalink**)

Originally Posted by okorop

great result eva!!! I think your ram with chip elpida are the best for intel core i7

Yeah with CapFTP's new found ADDR/CMD Bus VREF tweak, its alot easier to bench for single super pi 32M at >DDR3-2000Mhz speeds. With ADDR/CMD Bus VREF tweak disabled I can really get single Super Pi 32M consistently stable beyond 1024Mhz 7-8-7-20 1T. Now it's alot easier

ADDR/CMD Bus VREF = Enabled set at 100/100/100

@DDR3-2100Mhz 7-8-7-20 1T at 1.71v bios set (1.715v real DMM)

Haven't tried musical chairs with each GT module testing different configurations yet which I'll do next. So masking tape label each module as #1, #2 and #3 and test all combinations in dimm slot 2+4+6.

1/2/3
1/3/2
2/3/1
2/1/3
3/1/2
3/2/1

19-03-2009, 11:13 AM #7 (**permalink**)

Originally Posted by eva2000

Yeah with CapFTP's new found ADDR/CMD Bus VREF tweak, its alot easier to bench for single super pi 32M at >DDR3-2000Mhz speeds. With ADDR/CMD Bus VREF tweak disabled I can really get single Super Pi 32M consistently stable beyond 1024Mhz 7-8-7-20 1T. Now it's alot easier

ADDR/CMD Bus VREF = Enabled set at 100/100/100

@DDR3-2100Mhz 7-8-7-20 1T at 1.71v bios set (1.715v real DMM)

Haven't tried musical chairs with each GT module testing different configurations yet which I'll do next. So masking tape label each module as #1, #2 and #3 and test all combinations in dimm slot 2+4+6.

1/2/3
1/3/2
2/3/1
2/1/3
3/1/2
3/2/1

great work

push hard

12:36 PM

To add to this, but not to take away from eva's blazing speeds I wanted to share a little bit I've been doing with these Kingstons HyperX by tightening the timings to match the specs of Mushkin Redline rated at 1600 cas 6 timings.

The Mushkins will run at 6-7-6-18 1.65v, so far I've been reluctant to use them for super pi or WPrime benchmarks because for some reason they perform a second worse on each test in comparison to some Cruicial Value ram I overclocked from 1066MHz to 1600Mhz 8-8-8-21. That seems sort of odd, but even more strange is that EVERY application on my machine functioned the opposite way. Things were much more snappy. Higher CPU scores in vantage and a minor increase in frame rates. The only problem I'm finding with the Redlines is that they are unreliable... 4 days of owning the sticks and they are exhibiting signs of degradation. I saw the Kingston HyperX on newegg.com for $240.00 so I had to buy.

Here's what happened when I underclocked and tightened

My system is running at 4.2GHz [HT ON] 1.4000 Vcore in bios 1.408V from the multimeter, QPI @ 1.37V, DRAM @ 1.65V, Nominal VTT PWM frequency, -30% DRAM PWM frequency. Like it says in my MSN conversation I've NEVER had my i7 complete a 32M run. Everything else would be rock solid, but the XS Mod would always crash at around loop 15-18. Motherboard used was my DFI x58 T3eH8. Like I said earlier this is to show how flexible this ram can be and I am amazed. I hope this adds to the info eva has been able to provide so far.