Need some advice with first overclock

[1Wolf]

So...I've been reading and learning for quite some time now...and have finally decided to try my hand at overclocking.

Using Graysky's guide as my bible (If you happen to wander across this thread, Thanks SOOOOOOO much for the time and effort you put into that guide...it REALLY helped me to understand this process better!!) as well as some good 680i/780i/Q6600 guides here and there on the net...I set to work.

So far, so good I think...and I'm at the stage where I'm going through long periods of stress testing with Prime95. I've been collecting good data and writing everything down.

What I'd like to get some advice on...is whether you guys think I'm on the right track and I should keep trotting along this path? If so, some advice on things to try next or plans of attack would be great. Or perhaps you think I've interpreted my data badly and I'm coming at this all wrong.

First off...here is my rig:
Intel Q6600 @ 2.40 Ghz (B3 Stepping)
ZEROtherm Nirvana
EVGA Nvidia 680i SLI A1 Motherboard (Have NOT done VDroop mod)

Goal: Overclock to 3 Ghz (1333 FSB)

Next...a note on temperatures...

If you're an experienced overclocker and you're reading this, you probably already know full well that as a newbie overclocker with a Q6600 B3...I've had a hell of a time figuring out what my 'correct' temperatures are. As I've googled and searched and read and read....it seems last summer....folks relied on Coretemp and felt these chips ran really really hot. More recently, it seems the value for TjMax became the subject of debate, and difficult to nail down as Intel doesn't publish it....so folks have started to rely on the newer Real Temp to produce more reliable temps. The difference in my temps from CoreTemp (Defaults to TjMax of 100) and Real Temp (Defaults to TjMax of 85) was 15 degrees.

My cores were running low-forties idle and nearly 60 load reported by Core Temp...or....low-thirties idle and mid-forties load reported by Real Temp. I posted here on [H] for help and folks here seemed to tend to believe that Real Temp was more accurate...and that I needed to make sure I calibrated it properly. So I followed the documentation posted and think I did a pretty good job of calibrating Real Temp.

So...in my 'data' below...I'll include BOTH the Core Temp (TjMax 100) readings AND the calibrated Real Temp (TjMax 85).

Leaving all my voltages set to "Auto" adjusted them and allowed me to POST and load into Windows XP just fine. So I began iterating through graysky's process...of lowering core voltages one at a time, stress testing with Prime95, and collecting data.

Allowing the BIOS to choose my voltages set them as follows:
CPU: 1.40
FSB: 1.4
Memory: 2.10 (Manufacturer's suggested setting)
NBridge (SPP): 1.40
SBridge (MCP): 1.500
SPP <> MCP (CPU FSB): 1.20

This POSTS and boots into windows.

Starting with VCC...I started trying voltages on the "Low" side. The first "Stable" voltage that would POST and boot into XP was:

CPU: 1.36250 (CPUID reads 1.320)

CPUID reports voltage drop to 1.288 under load

Core 4 failed after 17 minutes Prime95

So I increased CPU voltage and tested again:

CPU: 1.36875 (CPUID reads 1.328)

Core 4 failed after 1 hour and 44 minutes Prime95

At 1 Hour of Prime95:
CPUID reports voltage drop to 1.296

The same core failed, but seemed to be more stable than before...so I increased voltage again (This is the point at which I started recording temps with my data, so I'll include them below):

CPU: 1.37500 (CPUID reads 1.328)

Core 4 failed after 1 hour and 9 minutes Prime95 (Faster than last run)

At 1 Hour of Prime95
CPUID Voltage reads drop to 1.296
Room Temp: 25
Real Temp Load Temps: 50 48 51 48
Core Temp Load Temps: 65 62 66 61

I wasn't sure what to make of the same core failing 'faster' under more voltage. So I bumped the voltage another notch and tried again. However, I didn't know if maybe it failed 'faster' because I needed to add voltage somewhere else. So I bumped several by just one notch. I figured that I could work backwards...and if I could get it stable for a long period of time in Prime95 with higher voltages...then I could work backwards and lower them one at a time.

CPU: 1.38125 (CPUID reads 1.336)
SBridge (MCP): 1.525
CPU FSB (MCP <> SPP): 1.25

Core 4 failed after 1 hour and 53 minutes

At 30 minutes into test:
CPUID reads voltage drop to 1.304
Room Temp: 25
Real Temp: 48 46 49 46
Core Temp: 66 62 67 62

At 1 hour into test:
CPUID reads same 1.304 voltage
Room Temp: 25
Real Temp: 51 48 51 48
Core Temp: 66 62 67 62

At 1 hour 30 minutes into test:
CPUID reads further drop to 1.296
Room Temp: 24
Real Temp: 51 48 52 48
Core Temp: 66 62 67 62

Core 4 fails 1 hour 53 minutes into test:
CPUID reads same 1.296
Room Temp: 24
Real Temp: 52 52 47 49
Core Temp: 72 67 72 65

Same core failed again, core 4, but at least it went longer this time...so I thought perhaps I'm on the right track. The 'heat' was starting to scare me though....trusting Real Temp was like a leap of faith. I saw the Core Temp numbers pop over 70...and thought "EEEEEK!" (well...it wasn't eek...it was another exclamation entirely). Real Temp numbers were still under 60 though...so I hoped I could keep trusting them. The PROCHOT indicators never lit in the Real Temp window...so I was hoping that meant Real Temp was indeed the accurate reading.

I bumped the CPU another notch, and left everything else the same.

CPU: 1.38750 (CPUID reads 1.344)

Core 4 failed after 2 hours and 46 minutes

At 30 minutes into test:
CPUID reads drop to 1.312, then drops further to 1.304
Room Temp: 23
Real Temp: 50 51 48 47
Core Temp: 65 66 61 61

At 1 hour into test:
CPUID remains at 1.304
Room Temp: 23
Real Temp: 52 52 48 48
Core Temp: 67 67 62 62

At 2 hours into test:
CPUID remains at 1.304
Room Temp: 23
Real Temp: 54 54 50 50
Core Temp: 69 69 65 65

At 2 hours and 30 minutes into test
CPUID drops further to 1.296 and fluctuates just a bit to 1.288, but stays at 1.296
Room Temp: 23
Real Temp: 60 60 56 56
Core Temp: 75 75 71 71

At this point, I decided to pull the side-panel off the case to make sure there was
plenty of airflow with all this 100% load testing...as Real Temp was getting close to my own 'heat limit', and Core Temp had passed it long ago.

At 2 hours and 46 minutes, core 4 failed
CPUID remains at 1.296
Room Temp: 23
Real Temp: 57 57 53 53
Core Temp: 72 72 68 68

Still Core 4....still POSTS fine and runs Windows fine...so I thought maybe that meant I should bump the CPU one more notch.

CPU: 1.39375 (CPUID reads 1.352)

Core 4 failed after 2 hours and 29 minutes

At 30 minutes in:
CPUID reads drop to 1.320, then another drop to 1.312
Room Temp: 23
Real Temp: 48 45 48 45
Core Temp: 61 58 62 57

At 1 hour in:
CPUID remains at 1.312
Room Temp: 23
Real Temp: 48 45 48 45
Core Temp: 61 58 62 58

At 1 hour 30 minutes in:
CPUID remains 1.312
Real Temp: 48 46 49 45
Core Temp: 62 58 63 58

At 2 hours in:
CPUID reads drop to 1.304
Real Temp: 50 48 50 48
Core Temp: 65 61 65 61

At 2 hours 29 minutes in, Core 4 failed:
CPUID reads further drop to 1.296
Real Temp: 57 53 57 53
Core Temp: 72 68 72 68

So...thats where I'm at now...

I'm not sure if I'm on the right track...and if I add another notch or two to the CPU, it will become more and more stable...and if thats the case...I just need to add notches until I can get it stable for 24 hours....or...until I get to 1.45 volts (at which point I really shouldn't go any higher).

Or....maybe...is this what processor "Throttling" looks like? Prime95 drives my temps higher over time....and it seems my voltage drops further the longer Prime95 runs and the higher my temps...and then Core 4 fails. if this is what "Throttling" looks like...then that would tell me that Core Temp was pretty accurate all along...and that it was Real Temp that was provided a reading far cooler than it really is...even after calibration. I hope thats not the case

So...I'm a bit nervous about continuing until I get a little input here. Or maybe I should just back down the overclock and 3 Ghz on a Q6600 B3 on Air is just too ambitious.

Your thoughts and advice would be greatly appreciated...and thanks kindly ahead of time for reading through all this junk and trying to make heads or tails of it.

 

[BillParrish]

until I can get it stable for 24 hours

I get bored if I play Prime95 for more than an hour. Use your game or primary application for long term reliability testing.

Go by the option setting (available in both programs ) worded like "show delta to Tj" that will tell you exactly the raw data from each cores DTS and will tell you exactly how much thermal headroom (deg C until the core thinks it is to hot, its like a countdown timer)

What is interesting is (assuming you listed them in order, core 1 core 2 ... and I am pretty sure you did) core 4 is not any hotter than the others, in fact most of the time cooler. This is why I say test with something else. It appears core 4 is weak and there may be nothing you can do about it. Continuing to pour on the voltage is not something I am fond of but you might have no choice, try up (slowly) to 1.425 but use something else to test. If you are a gamer loop 3Dmark after one hour of prime and see what happens. If what you use the computer for will run for a week solid, who cares how long prime 95 will run.

 

[1Wolf]

Thanks Bill

Go by the option setting (available in both programs ) worded like "show delta to Tj" that will tell you exactly the raw data from each cores DTS and will tell you exactly how much thermal headroom (deg C until the core thinks it is to hot, its like a countdown timer)

Terrific. That'll help greatly as I've just been nervous about this whole temp thing. Its a pretty big temp difference between the two programs, using one program I'm "fine" under load...using the other program...I'm over the intel 62.5 spec on my B3 Q6600. This being my first overclock, and not knowing 100% for sure which temps are correct...its a nail biter.

This is why I say test with something else. It appears core 4 is weak and there may be nothing you can do about it. Continuing to pour on the voltage is not something I am fond of but you might have no choice, try up (slowly) to 1.425 but use something else to test.

To make sure I understand this correctly...are you saying that I should NOT use Prime95 to test (beyond 1 hour) because, due to my "weak" 4th core, Core 4 would most likely just continue to fail in P95 no matter what I might do with voltages and what not?