E8400 with P5K-Pro can't get 500FSB Stable, HELP!

[sunderstan]

Hey guys I got some overclocking stability issues for a period of time with this CPU and board...
Spec:
M/B: ASUS P5K-PRO BIOS 1101 (Also tried 0601, 0806, 0904, 1001, 1002, 1101)
CPU:C2D E8400 (Q807A191)
RAM: G.Skill CL4 2GBPK (PC 6400) DDR2-800
HDD: WD5001ABYS 500GB
Display: Galaxy 8800GT
PSU: Delta Server Power 550W
NB Cooler: HR-05
CPU Cooler: PC Cooler Southsea W 7000RPM 12cm FAN (Ultra Noisy)

Settings:
Multiplier: 6
FSB:500
FSB Strap to Northbridge: 400Mhz
FSB:RAM Ratio: 1:1 Sync
PCI-E Freq.:104Mhz
RAM Timing: 5-5-5-15-3-40-6-3-3
Static Read Control: Enabled
Transection Booster: Disabled with Relax level 0 (Which Gives tRD 9)

Voltages:
VCore:1.3v
CPU PLL:1.50v
FSB Termination:1.34v
vDIMM: 2.2v
Northbridge: 1.45v
Southbridge: 1.05v
Clock overcharging: 0.8v
Loadline Calibration: Enable
CPU GTL Ref. voltage: 0.61x
Northbridge GTL Ref. voltage: 0.61x
CPU Spread Spectrum: Disable
PCI-E Spread Spectrum: Disable

Plus: All CPU Features like C1E and Speedstep are DISABLED

Here is the case...
I was trying to get to 500FSB with this board however failed from time to time
I can never get stable above 480FSB as memtest failure and frequent BSODs
My board can get 480FSB Stable with northbridge voltage @ 1.25v
however no luck on going higher FSBs. In memtest I keep failing #7 (Random number sequence)
with error bits 2000000, and sometimes 00000001 when FSB Voltage in faulty
Other tests just passed and works fine except #7
I've tried up the NB voltages to 1.55v but it seems only make thing worse with more and faster errors
I also tried loosen the RAM Timings to like 5-5-5-18-5-60-6-5-5 and it make things worse too
I have also tried only one stick of RAM and it pass all tests without any problems so I don't think memory is holding me back
I've done some hardcore research but seems the problem is pointed out to be the MCH issue.
Actually I have this board RMA once as it failed to POST after I replaced the Northbridge heatsink.

Do you guys having the same problem as me? Any suggestions?

 

[Avarice]

i would say you are pushing the ram too far try more voltage or get faster ram. faster ram would probably be the best idea in my opinion. also what is the voltage on the ram?

 

[Deleted member 24505]

try running the cpu speed lower and run the ram as close to what it would be at 500 fsb using the dividers to get the ram speed.You will know if its your ram at least that cant take the oc speed.

 

[CyberDruid]

Check to make sure your mobo is not being warped by the CPU mount: this will draw the chips away from the HS. Use a backing plate on the CPU mount to avoid this..once the board warps there will be an airgap between the chips an HS that needs to be pressed out with a piece of wood/foam/plastic w/e between the tray and mobo slightly thicker than the height of a stand off...

I also think your volts are too low on the chipset...

EDIT: use Orthos to determine if the RAM is the prob. No reason to run 1:1...if the RAM doesn't like a particular divider or strap (they all vary...certain chips hate certain straps) find the one it does.

TBH you will get better results running your board at a lower FSB if the higher FSB makes it fault: I have found this from long term Folding...you may think high FSB is going to get the best performance: but actually it is more to than that: you get the best performance when you have the fewest faults during calculation...and on the tightest strap that works with the RM...

FYI I can get my P5K deluxe to 530 and it will Fold at 8 x 500 RAM at 1254...but it Folds a lot faster at 9 x 444...RAM at 1138 and with a lot fewer voltage tweaks (less heat) and no crashes.

I also leave all my GTL settings stock/auto...

It's a classic argument: Tight Timings slower speed versus Looser timings higher speed: which gives better performance? Sometimes it's not about the speed but the timings...

 

[sunderstan]

Thanks for all your help first...
For the RAM Part, I actually have an AEXEA Ram PC8500 @2.2v
This memory module has been tested in CPU stock speed and FSB but with different divider and pass memtest without any errors overnight.
However, when I tried to use this pair for rushing 500 FSB the same problem happened again:
Memtest #7 Failure. The best that I can get is 6 passes with 3 errors...
A ram can run @1066Mhz that can't run @1000Mhz with 1:1 dvider seems hard to admit...maybe there's something to deal with the subtimings (Like tRFC, running @55)

And for the HS Part...yes, the CPU heatsink and the NB HS nearly sticked together but I don't think it's the major problem??...And also while running MEMTEST the NB is d**n hot...even can burn your finger. For the voltage part...I've tried 1.55v before but it's only getting worse and I'm running 1.45v now...I'll be there still attempting and will post again if there's something new...

Any more suggestions that I can do with the settings? Please...

 

[CyberDruid]

Get a spare fan and postion it or tie it so that it blows air right on hot parts: it makes a huge difference when stress testing...you might be able to idle all day...but the minute you load up it blue screens: that can be heat related.

Without a doubt dividers (straps) make a huge difference when pushing up the FSB. Sure Relaxing Transaction Boost to 3 can help..but then you lose speed...for whatever reason certain dividers are easier to CO. Regardless of what you want what you get is instability at 1:1. I really think that is a non issue for performance on the P35...choose a friendly divide that can handle a good CO on the RAM then use MemSet in Windows to tweak the subtimings

I have found that almost regardless of chip my P5K needs about .2 volts more than I think it should to run a high FSB. By High I mean anything above 475...It loves 400 (1600) but when I push to 500 and beyond I am maxing out the voltage.

I had a P5K3 that was VoltModded and that would push till forever...but running anyting over 480 was super hot due to all the voltage: I had the entire chipset and MOSFETs and VRM areas LCed...but it still had a limit.

The P35 is only going to get you so far on the FSB is my impression. Despite LN2 runs on the DFI P35 that posted 600+FSB the reality is that even with special cooling you are not going to run a 500 FSB stale. At least I have not figured out how.

 

[JrRacinFan]

Try loosening your GTL's a little bit, maybe up to .67 each. It may help.

 

[erocker]

Have you tried disabling static read control? x63 on the CPU and x67 on the NB may help also.

 

[sunderstan]

Yeah I've tried some crazy stuff just hours before...
I've set FSB to 533 and the machine was able to POST @ 1.45 northbridge voltage
what I've proved is 1:1 FSB-RAM ratio is totally different from 5:6.
With 2.2v vDIMM 5:6 @ 1066Mhz I was able to pass memtest loops including #7 without any errors
However with 533 FSB it won't even POST @ 2.2v therefore I pumped to 2.4v and it POSTed but memtest fails deadly.

And yeah. I've tried having static read control disabled and transection booster relax level 3 which gives around 13 tRD but it only cause more errors...I wonder if there is something to do with the combination of memory subtimings?
Another test: 1.32FSB 1.45NB with 0.63CPU 0.67NB GEL Ref. fails memtest #7 3 errors each pass...