C2D Q6600 FSB OC issue just started

G

GeminiCool

Weaksauce
Joined
Jan 11, 2005
Messages
97
I've been OCing for some time and I really believe the Intel CPUs tend to burn themselves out even when temp is kept under control. I'm asking this question to get some feedback.

I had a Gigabyte GA-P35-DS3R, I wound up giving it to a friend.. it's a long complicated story. On that board I had been running around 3.3GHz with 366Mhz FSB and a mem divider that kept me below 880Mhz. About 6 months ago I had to drop the mem divider down so that the memory would run under 800Mhz. Fed up with the tweaking I really didn't bother investigating past a memtest86+ for 24 hours without errors.

I replaced the Gigabyte with an Asus P5Q Pro Turbo about a month ago. With a fresh MB I decided to begin my OC process again.

I've memtested my 4 sticks of memory. They ran for 24 hours without a single error at 880Mhz 5-5-5-18 @ 1.84V.

Now, if I set the FSB NB strap to 400Mhz, the PC won't boot, period. if I set the FSB NB strap to 333 and set a memory divider such that I'm running around 850Mhz and about 3250Mhz on the CPU as soon as I start Prime95 I get a crash.

I have tested the CPU stability using boot utilities and the CPU is stable without Windows. But when I add 3.2Ghz + 850Mhz memory + Windows + Prime95 = Crash.

I've read bits and pieces about the NB timings and such, could there be something with the FSB and the NB timing as I increase the speed? My CPU voltage is at 1.39V, using the Corsair H2O cooler, with a full load prime95 Temp (when I'm stable) of 50-55C max.

I'm really at a loss here. I had a 1.6A P4 behave like this. After 3 years @ 2.4Ghz, the CPU would no longer tolerate a FSB OC. The only time it would fail would be in windows.

Has anyone else with an Intel chip experienced this degrade in reliability? I have not seen this on my AMD chips.. not trolling just telling it how it is... but then again my AMD chips only see a 10-20% OC.

Any ideas?
 
could there be something with the FSB and the NB timing as I increase the speed?
Click to expand...

Yes. And more than likely it is the cause of your discontent.

It seems you have a good understanding of the numbers and how the settings interact at least as far as the speeds go but are missing the big picture. In addition to raising the clock speed of the CPU and memory, at the same time you are messing with the speed of the data buss that transmits data and addressing information to/from the CPU to the MCH/Northbridge and to/from the MCH/Northbridge to the memory. The MCH/Northbridge sits between the memory and the CPU and as you likely gathered from your reading, becomes OCed just like the CPU. It can be considered a specialized CPU too. Increasing its voltage (and keeping it cool) helps it cope with higher frequencies and minor changes to the FSB voltage (Vtt) may also help. It is a very complicated system running on razors edge timings. The manuf should get a lot of credit that the CPU and board and memory are made so forgiving that most of the time you can slam in a higher FSB and it will work with a significant OC. However once you get past some speed that varies with each individual component combination you are going to have to dig deeper into the specialized settings to compensate for the effects a higher FSB speed has on the data traveling over that buss.

Up your MCH voltage first one bump, test, then another bump, retest and then one more if needed. Around 1.325V you will want to stop. Put a fan on it.

Then if still no joy, leave the MCH at the second "bump" of increased voltage and try increasing the FSB reference voltage (Vtt), not quite sure how it is labeled in your bios, a bump and retest.

There are likely other settings that might help. As I am not up on Asus bios I cant really help much more.

There are a few reports of OC's "going away" but not a lot. I attribute most of them to very high Vcore or changes to the motherboard due to heat and time. I myself have not seen the issue with quite a few Q6600 machines running at 3.2 built for customers over the last 3 years.


edit: rereading your post (it is a bit disjointed).

It is not uncommon for after about a month for the board to change its characteristics, part values change, thermal pads/grease cures etc. etc. A lot of us avoid that by heavy duty stress testing during the initial OCing.
 
Last edited:
How old is the power supply? What brand and rating does it have?

I've had a couple different power supplies blow a cap or two and cause me to have to drop back an OC on a system.

You may also want to rermove the NB and SB heatsinks, clean the stock junk off of there and apply some real TIM. That can help quite a bit.
 
Hey guys, thank you for your feedback.

On the PSU front, it's a Thermaltake 700Watt SLI & 80 Plus certified PSU. It's about 2.5 years old. I run an APC UPS with LCD, the UPS says my system, fully loaded, is actually under 300 Watts! This is a separate issue, but for 99% of the people out there, I'm thinking a quality 500Watt PSU is more than enough. I run the Q6600, a pair of 7200 rpm drives, and a 4870. I've watched voltages in HWMonitor and there is no major deviation.

On the MB front... My specific knowledge is limited. My understanding is that Asus opens an option called north bridge strap to allow the user to relax the timings to allow for higher FSB overclocking. Now my MB is a P45, the 333Mhz FSB should not be an issue for this board, yet it is. Infact, with the 'strap' set to 400Mhz which should be the most loose of all timings, I can't even boot when the Memory speed & FSB speed are in spec!

I guess here's my disconnect. If I set the FSB to 345Mhz I can run Memtest86+ for 24 hours straight without a single error. I can boot into Windows, run Prime95 stress test, but only the one which stresses the CPU, very little memory. As soon as I do the blend test.. bam, threads crash or the computer hard locks.

I messed a little with the NB voltage, never ran it above 1.3 for any amount of time. From 1.2 to 1.3 in steps of .2 it makes no difference. My case is a Cosmos S, and I game in the basement. The case has great airflow including a big side panel fan and the ambient temp is low, under 70F. Heat was not an issue during this testing.

I guess this all really points to the north bridge on the MB, right? I mean the CPU sees some FSB and that's that... The memory sees some memory speed and that's that... The NorthBridge connects the two, so if there are problems running the memory & CPU in spec with a NB speed of 333 or 400 then the motherboard is defective... is this thinking wrong?
 
What BIOS are you using? You might want to check out Ket's mBIOS for the p5q's, they have better memory support. General consensus says they are better for OC'n especially on quads.

The p5q's are hit n' miss with the straps, you're better off leaving them on auto and manually selecting the dram frequency if you have any problems. You're also using a memory divider, until you're stable run it at 1:1. Meaning FSB @ 345 set your DRAM frequency to 690, FSB@ 366 DRAM frq=732

Disable:
DRAM Static Read Control
DRAM Read Training.

In addition you may need to manually set the AI Clock Twister to lighter settings and AI transaction booster should be fine on auto. Play with it AFTER you get everything else stable lowering it to 10-13.

You don't address the PLL, FSB(vtt), or DRAM voltages. The p5q's notoriously overvolt just about everything also.

CPU Voltage: what it was stable at on the gigabyte+1-2 notches (lower it after you reach stability and test. my cpu's need more juice in my p5q then my DS3's.
CPU GTL Reference:Auto or 0.65x
CPU PLL Voltage: 1.52-1.56
FSB Termination Voltage: 1.22-1.26
DRAM Voltage: your specs maybe +.1
NB Voltage: 1.22-1.3--test up to 1.35 , p5q's are a lil' hungry here especially with 4gb ram or more.
SB Voltage: auto is fine or 1.10
PCIE SATA Voltage:auto is fine or 1.50
Load Line Calibration: Enabled
CPU Spread Spectrum: Disabled
PCIE Spread Spectrum: Disabled
CPU Margin Enhancement: Performance Model
 
I don't get it... I promise I tried nearly every setting... The only things different from what I'd done prior and Loboking's advise is the FSB Strap, CPU GTL, CPU PPL, and CPU Spread Spectrum... is auto really that far off?

Loboking - Thank you. In disbelief, I set every setting, started at a speed of 350, went up to 400. At 400 the CPU became unstable, I didn't want to push more than 1.4 Volts. I backed it down to 380 and Prime95 ran without issue for 8 hours.

I'll play with it a little more over the next few days... it's hard to believe such a minor tweak (a change from auto to near min values) could have had such an impact.
 
The p5q's overvolt everything on auto. Just setting all relevant voltages to a couple notches past stock/auto is almost a necessity to have voltages in spec.

These boards are great OC's but they are finicky and have a couple quirks compared to the gigabyte's.

Yes, for any overclock you want to turn off all Spread Spectrum for stability and to lower the strain on the mobo. You can try with it on AFTER you get stable.

Glad i could help! I had a couple growing pains when i got my p5q to replace my ds3.

Before i forget. About your memory frequency and AI transaction Booster. Once your stable at whatever speed you choose you can try using a mem divider to run your ram closer to its rated speed to get a little more out of it.

Lastly I would strongly recommend you set AI Transaction Booster to manual and a common performance level as low as you can get stable with. This option controls the tRD, the lower this is the faster and more responsive everything feels. Auto is always to high ~ 14-16 usually for me. Try 10-13 but try to get stable lower, I've seen successful OC's running at 6-9 CPL (common performance level) and that makes the OS just pop!

edited this after I realized you're not using ddr800 and to cover an embarrassing math error.
 
Last edited:
You must log in or register to reply here.
Back
Top