How to get the volts down?

[ehume]

I'm trying to OC the Hotrod rig in my sig. It will run 4GHz eay as pie with MB defaults, but that sends voltage to 1.396v, and with a Megahalems and a San Ace 9G1212H101 the temps play at 80c on load (OCCT doing Linpack).

Since my ambient is a 17c basement I need a cooling solution that maxes at or below 67c so I can run it upstairs at 4GHz without worrying about ambients (I don't expect greater than 30c).

I would like to run this machine at Vcore=1.25v. I've tried Vtt at 1.19v, PLL at 1.9v. I've given it a skew of 200ps. But OCCT still gives me an error at the end of the second rep or in the middle of the third.

I'd appreciate some tips.

 

[NumericalMethods]

What base clock and multiplier are you using?

Have you tried OCCT small ffts for 1 hour? Does it pass?

Have you tried OCCT large ffts for 1 hour? Doest it pass?

For the voltage and speed you are running that temperature looks par for the course, especially with HT on.

 

[ehume]

Sorry for the missing data. I've tried 200x20, 191x21 and 182x22. I'm working on the 182x22 now, hoping that the lower BCLK will require less volts. HT on, because I will be running this rig with HT on. I'm using OCCT's standard Linpack setting because it is standard. Earlier I used LinX with 768MB blocks, and I was able to work with 1.25v. I'm using the OCCT version of Linpack because it maxes the system out - I'll know that there is nothing I can do that is more strenuous than that.

I alos know that every hair I increase Vcore will greatly impact the cpu temps. Hence the effort I am putting in to get that down.

 

[NumericalMethods]

Ok cool. I think the low bclk higher multi is a good strategy for minimizing temperatures.

The way I dialled in my settings was to first drop the CPU multi low and concentrate on the bclk. I used OCCT large FFTs to do so. This is far more stressful for the memory controller and RAM than Linpack despite what people say about Linpack.

Once I got my bclk fairly stable - around 2 hours OCCT I then worked on raising my CPU multi and voltage. In this case I strongly agree that Linpack is the most stressful thing you can serve your CPU. I raised my voltage high enough to get into windows and then started to drop it down so it passed 3 passes. When it started failing 3 passes I knew the voltage was marginal. I then raised the vcore 2 increments and tested for 100 passes for stability. Then I went back and tested my bclk again to check the OC hadn't effected things.

I too came up with different running scenarios for bclks and chose the one that fitted performance/ volt best - 190*20 for 3.8GHz. I found my chip needed 1.32v for 3.8 and 1.42 v for 4! So the choice of final running speeds was clear.

I suggest you try something like this to work out what your practical maximum 24/7 will be. Good luck and report back with how you get on

 

[Kenmitch]

Hmm....Making your goal 4ghz with HT enabled and trying to keep your temps down is maybe unrealistic.

To me it seems like your options are limited to getting a better cooling solution or lowering your overclock to keep the temps inline.

You can only do so much playing around with the voltages. I'm sure that on auto your MB is feeding more voltages than needed to run at set speed. But if it's stable on auto then those set voltages would be your best place to start when you take manual controll of them. I'd suggest booting on auto and if stable go back into bios and make notes of all the voltages that the board feeds it.

Are you scared of the cpu vtt 1.21v max? I ask as it is pretty much the make or break voltage with these chips.

 

[aamsel]

I have a similar setup here, with an i7-860 and an ASUS P7P55D Deluxe.
I think your hopes have proven to be unrealistic.
All you can do it to throw more voltage at it, and the temp will go up.
The secret to some of these Nehalems is to turn HT off.
I did it with my 860 and I am at 184 x 22 stable for a 4.052GHz, again, with HT off, and all
the INTEL and other extras completely turned off.

I am at 1.31v with load line calibration off (no anti-vDroop correction),
and again, almost everything else turned off.

HT adds enough extra heat to crash this same setup to this point.
I plan to have "FULL POWER HT OFF" saved in the BIOS, as well as a
"REDUCED POWER HT ON" or somesuch. It will pass the same settings
with load line calibration on, but the heat is higher than I want.

 

[ehume]

NumercialMethods - what did you vary when you were stabilizing your memory?

I have memory that is rated at 2000MHz 9-9-9-24. I've had it there at those settings. But with various studies showing no benefit to fast memory I've been running it at 1600MHz and lower. But still, since I'm not usually getting BSOD, but only an Error! message by OCCT, I'm wondering if there is a memory issue.

Kenmitch - I fear you are right about the heat and the cooling so I bought a Noctua NH-D14. But still, I'd like to squeeze what I can out of the Megahalems. After all, this most recent heatsink review found the Mega to be the best, significantly ahead of the D14.

And yes, despite Miahallen recommending +0.2 to +0.3v for Vtt, I am staying under 1.21v. In fact, when I just tonight nudged up my Vcore to 1.2625v I reset all the other voltages to Auto.

aamsel - thanks for the specific tips. HT Off may be the only way I can hold my temps down. With my previous MB I was up above 1.3v and had no problems with stability, but the cpu ran too hot. But then, I had HT On then.

 

[ehume]

I just retrieved today's notes from the basement. Ambient = 16.5c.

Vcore set to 1.25625v. All else auto. BCLK 182x22=4004. Turbo Boost auto. All other heat saving disabled.
Skew = 0. Memx8=1456. QPIx32=5.2GHz, uncorex18=3276. HT on. LLC auto.
Testing 2 CM Blademasters on PWM. HST=57c, CPU0=67c
Vcore reported 1.248v on idle, 1.264 on load. Dips to 1.248v. No spikes.
OCCT set for 18min. Error at end of 11th rep.

Reboot. skew 250ps. all else same. Error after 2d rep.

Reboot. skew = 0. cpu 191x21=4011. Turbo disabled: maybe the cpu was trying to boost a single core.
In eighth rep, went to 50% cpu usage, then error message. Vcore dipped to 1.248v. No spikes.
HST=59c, CPU0=68c. So the higher BCLK made a difference.

Was the first setup actually better?

 

[Kenmitch]

And yes, despite Miahallen recommending +0.2 to +0.3v for Vtt, I am staying under 1.21v. In fact, when I just tonight nudged up my Vcore to 1.2625v I reset all the other voltages to Auto.

Not to scare you or anything but most MB increase all the voltages when you start playing around with overclocking. So you might wanna see what it's feeding the cpu vtt for voltage to be safe. My MB feeds 1.35v on auto if I remember correctly.

I'm not sure if it's a myth or truth but I read these chips overclock better when using an odd multiplier such as 21x etc. I'm using my 21x 4 core turbo multiplier as I'm shooting for max what I feel safe at overclock....Although I did do a auto voltage bootup into windows at 4.597ghz for the screenshot

I'm on water so don't have to worry about the heat issues...Unless I wanna run at 4.4ghz or more anyways

I'm running my i5 750 @ 4.2ghz (200x21) I have C1E,Turbo and speedstep enabled still. Yep I have to very slightly bump my cpu vtt above the spec to 1.26 to be 100% stable. I figure it will be safe and cause no problems at all.

I use Intel Burn Test and the evga eleet program to tweak my voltages. IBT will error out very quickly. The eleet program allows me to change voltages on the fly but I always just set the block and multiplier in bios. Once I figure out my voltages I just set them in bios.

I also use IBT to fine tune the voltages also. Just gotta keep a note of the gflops as it's possible for the system to be stable but not optimized I guess you'd say. Depending on the chip I've seen anywhere from 2-6 gflops diff by just tweaking the voltages just a little bit more....Small increase with vcore cpu vtt the biggy on this.

I'm kinda getting good with this board and most likely could get any chip within 5% of max overclock within minutes....Not counting the time it takes to put the chip in the MB of course

 

[NumericalMethods]

NumercialMethods - what did you vary when you were stabilizing your memory?

I have memory that is rated at 2000MHz 9-9-9-24. I've had it there at those settings. But with various studies showing no benefit to fast memory I've been running it at 1600MHz and lower. But still, since I'm not usually getting BSOD, but only an Error! message by OCCT, I'm wondering if there is a memory issue.

snip

I left my memory below spec with it's full rated voltage so I knew memory wasn't an issue initially. When I had my three different options of bclk I started looking at what my memory could do in terms of frequency and timings.

I found that command rate had a fair effect on performance - by fair I mean a minute 1-2% increase in performance in some stuff. My three configurations were 1448 7-7-7-24 CR1 @ 1.65v; 1520 8-8-8-24 CR1v @ 1.65; 1600 8-8-8-28 CR2 @ 1.65v.

Third set of timings were the ones specced for the RAM and performed the worst. Configurations 1 + 2 were approximately equal in performance. All three configurations needed a 0.05v bump over spec to run stable with the increased frequency on the IMC.

It will depend on your board, chip IMC and RAM as to whether it will run to spec speeds at spec voltage. I have seen reviewers successfully hit the advertised speeds sometimes at advertised voltage, sometimes requiring slightly more voltage to do it. HTH

 

[aamsel]

I solved most of my issues with the P7P55D Deluxe. I am thinking about trying a Xeon version of the i7-860.

Also, is this P7P55D Deluxe the P55 board to go with right now, otherwise which one is?
I still have return privileges.

 

[ehume]

I would like to thank you all for your help and your encouragement. The minimum setting that provides a stable BCLK=182x22=4004MHz overclock with my MB and my cpu is Vcore = 1.26875v, with all other voltage settings on auto. One sliver down gave me errors on OCCT, even with the Vtt set to 1.19v. I guess Vcore really is king.

With this setting, Vcore at idle is 1.248v. At load it is 1.280v and between max loads during the runs it is 1.264v.

Moving from Vcore set to 1.26250v added a degree to the heatsink spreader temp and CPU temps. The CPU temp above ambient went from 49c to 50c with twin 2000rpm fans.

Complete settings: Vcore set to 1.26875v. All other voltages set to auto. LLC enabled. HT enabled. BCLK = 182 MHz. CPU multiplier = 22x. Turbo = auto. Various power-saving features disabled (no C-states, etc.). QPI ratio = 32x. Mem multiplier = 8x.

Once again, thank you all.