[jimmsch]

So, I am getting a quad itch. I see some Q9650's (BX80569Q9650) on eBay for $250 free shipping. I was thinking of putting one in my main rig (see sig).

I really only wanna do this if there is a good chance...a GOOD chance, not just a chance...that it will reach the same FSB (445x9) as the E8400 that is currently residing in there.

Just curious if it's worth it. I do some video encoding and photoshop stuff, but not alot anymore.

So, four questions, really:
1 - Is $250 shipped for a Q9650 listed as new, processor only, no fan as good a price as it seems to be?
2 - Does the model # (BX80569Q9650) tell us anything about the stepping?
3 - Is there a GOOD chance it will reach 445x9 on my Gigabyte mobo?
4 - Will there be much of a performance gain from the current E8400 and the quad at the same clock when encoding video?

Thanks -Jim

[SeasonalEclipse]

From what Iv read the 9550 is alot better then the 9650 for the money. I bought a new one off new egg and a 775 board to rebuild my dead main machine Friday. Finally decided to go c2q since AMD has let me down.

[Boulard83]

Originally Posted by jimmsch

1 - Is $250 shipped for a Q9650 listed as new, processor only, no fan as good a price as it seems to be?

R : Its a good price for a Q9650

Originally Posted by jimmsch

2 - Does the model # (BX80569Q9650) tell us anything about the stepping?

R : ALL Q9650 are EO stepping.

Originally Posted by jimmsch

3 - Is there a GOOD chance it will reach 445x9 on my Gigabyte mobo?

R : I think you might be able to reach it.

Originally Posted by jimmsch

4 - Will there be much of a performance gain from the current E8400 and the quad at the same clock when encoding video?

R : YES

[jason4207]

^^+1

[jimmsch]

Well, I went ahead and did it. I will make a thread with some results in the coming weeks.

Thanks guys

[retrogreq]

ouch..you could have gotten an q9550 for less than 180 bucks from microcenter....and they are almost the exact same thing, especially when overclocking

[jimmsch]

I guess the keyword there would be almost. From what I've read the 9650 will almost definitely do 4ghz while the 9550 is a crap shoot getting that high. I suppose it is the extra .5 multi? Anyway, I don't see the point in upgrading to a quad if it means I have to run a lower clock.

I have been running at 4ghz ever since I put an X6800 under phase. When the E8400 came out and I managed to get it to 4ghz on air using less than 1.28 vcore, I got rid of the phase and have been using this for more than a year. I need my 4ghz, and I need it to be almost a definite. The 9550 is a possibility, but the 9650 is a better bet.

All that being said, the nearest Microcenter to me is close to 100 miles away.

[jimmsch]

I double posted to move thread to the top. Otherwise I would have started a new thread.

So, I received my Q9650 and started the process of overclocking.

I first set the mobo to fail safe default and loaded Windows. All seemed well.

Then, since I was stable at my settings using the E8400 for well over a year I set my BIOS to the saved settings for 4.0ghz. Figuring the RAM should be fine since it was stable at these settings for well over a year I just paid attention to the CPU.

My stable settings using E8400. These were rock solid for over year. The RAM is set to 2.00A (1:1) using 1.90vdimm.



On the Q9650:
I had the Vcore all the way up to 1.33 and still error-ed out of P95 (64bit) after just a few minutes. The temps got up to 65 on Core 0 and low 60's on the rest of the cores. The more Vcore I give it, the faster it seems to error.

Raising MCH always seems to result in no POST.

Dare I go higher on Vcore?
Should I go all the way back to defaults and start the whole process over?

My settings: Q9650

Multi=9
FSB=445
PCI-E Freq=101
RAM link=2.00A (1:1)
RAM timings=manual 5-5-5-15

RAM is rated for these timings at 800fsb using 1.8v. I run it at 890 using 1.90v. Was always stable at this setting for well over a year with E8400.

All voltages set to normal except for RAM which is set to +.10 and Vcore which I tried all the way up to 1.33.

The Q9650 VID = 1.2375

Any ideas/suggestions?

EDIT:
I am running BIOS F8, and I see that the Q9650 isn't officially supported until BIOS F9i. This is a beta BIOS...I hate beta BIOSes, but I suppose I have to try it. At least this mobo allows for dual BIOSes installed.

EDIT #2:

Working my way up the ladder on BIOS F9I.
Running P95 blend test for at least ten minutes at a time. Going up ten mhz at a time. Have yet to need more than stock Vcore (1.2375). Currently at 375x9 (3.375ghz). Load temp at 58 for hottest core.

[jason4207]

Get the right BIOS on there first, and then start from scratch w/ the multi low (6). Try to focus completely on the FSB at first (so keep the RAM slow and loose as well).

Make sure to never use that old profile after you update the BIOS. It won't be any good anymore as profiles usually don't work right across different BIOS'. If you want to save your settings write them down, re-enter them, and then overwrite your old profile.

If more vcore is making it worse I can assure you even more won't be any better. Something else is not set right.

Post all your BIOS settings for critique if you like. There are definitely other settings besides the ones you listed that are critical to getting it stable.


Also, while testing for FSB speed I've found that Prime95 large-fft is the fastest indicator that the NB isn't stable.

Once the FSB is stable it should be easy to up the CPU multi and find the perfect amount of vcore.

[jimmsch]

Thanks for the sugestions, Jason.

How long do you suggest to run the large FFT's while going up the FSB ladder? I obviously don't want to let it run for 8+ hours for every 10mhz increase in FSB.

Thanks again

[jason4207]

No just go for 5mins or so. Once you start failing in less time then that then start making BIOS adjustments and see what helps and what hurts.

I also suggest taking all voltages off of auto and setting them manually to the lowest voltage (except RAM and vcore). Leave GTLs on auto to start, but definitely play w/ NB GTL when you are trying to get the NB stable...you may have to re-try GTL settings after you make further voltage adjustments b/c they are interdependent. You can try all GTL settings if you want. They are not applied voltages, but just how the signal is read.

PLL, VTT, vNB, and NB-GTL should be the main settings you'll need to mess with, but there may be others.

You'll want to run Large-FFT longer as you finalize the OC. I like to go at least overnight, and if I have work the next day I just let it keep running until I get home.

[jimmsch]

OK, I have been still running up the 9x multi. Currently at 420x9 (3.77Ghz) still under stock Vcore and 1.90Vdimm.

I don't see any settings for VTT or GTL or PLL anywhere in the BIOS. On the next reboot I will write down all of the listed voltages available.

I have been running 5 minutes each of small FFTs, Large FFT's and Blend Test since Jason mentioned that Large FFT's error faster than the others. I figured I would give a few minutes of each.

And, yes, when I reach my desired OC (4Ghz) I will let the test run for at least 8+ hours.

[jason4207]

Large fails fastest if the NB/FSB is the problem...that's why I'm trying to get you to focus on the FSB.

You're way should work fine too, but I've always found that focusing on 1 area at a time (CPU, NB/FSB, RAM) allowed me to see how the settings affected everything better. Taking short cuts early can lead to frustration later, but that's just my experience w/ the limited hardware I have tried.

Have you hit cntrl-F1 once you get into the BIOS? It will seem like it did nothing until you go into the advanced (or whatever Gb calls it) menu and then you'll see more options.

I'm not familiar w/ that particular board and all it's voltage options, but I figured that since my P35 had those settings that your X38 would as well. They may be called something a little different.

[jimmsch]

I didn't know about ctrl+F1. I will try it at next boot. Thanks again.

OK, I found the missing voltages using the ctrl+F1, thanks Jason.

I found 420x9 to be at least 5 minutes stable onall three of P95 tests at stock (1.2375) Vcore.

Here is a list of my OC settings. This is in the main rig in my sig. RAN is rated for 5-5-5-15 at DDR2-800 1.8v.

Or 5-5-5-15 DDR2-1000 at 2.1v.
------------------------------------------------------------------------------------------------------------------------------------------------
Robust Graphics Booster = FAST
CPU Clock Ratio = x9
Fine CPU Clock Ratio = 0.0 (greyed out...is available at lower multis)
CPU Frequency = 3.95GHz (440 x9)
CPU Host Clock Control = Enabled
CPU Host Frequency = 440
PCI-E Freq(Mhz) = 100
CIA 2 = Disabled
Performance Enhance = Standard
System Memory Multiplier = 2.00D (1:1)
Memory Frequency = 880
DRAM Timing Selectable (SPD) = Manual = 5-5-5-15

Advanced Timing Control
Act to Act Delay = Auto (3)
Rank Read to Write Delay = Auto (3)
Write to Precharge Delay = Auto (6)
Refresh to Act Delay = Manual (42)
Read to Precharge Delay = Auto (3)
Static tRead Value = Auto (8)
Static tRead Phase Adjust = Auto (0)
Command Rate (CMD) = Auto (2)

Clock Driving and Skew Control
CPU\PCIEX Clock Driving Control = 800mV
CPU Clock Skew Control = Normal

System Voltage Control
Manual
DDR2 Overvoltage Control = +0.10
PCI-E Overvoltage Control = Normal
FSB Overvoltage Control = Normal
(G)MCH Overvoltage Control = Normal (it seems whenever I change this it results in no POST)
MCH Reference Voltage Control = Normal
DDR Reference Voltage Control = Normal
DDR Termination Voltage Control = Normal
CPU GTLREF1 Voltage Ratio = Auto
CPU GTLREF2 Voltage Ratio = Normal
Loadline Calibration = Enable
CPU Voltage (Vcore) = 1.2375

I also have disabled things like EIST, C1E, onboard audio, one of the LAN ports and anything else I am not using atm.

Will run the three tests of P95 for 5 minutes each at the above settings.

All tests crashed in less than one minute. They did good at 420 FSB. I will raise Vcore one notch.

Running the above settings with 1.24375 Vcore...one notch up from stock. Less than a minute crash small FFT's
WIll try another Vcore notch until I get further instructions.

Small FFT's failed at 1.25000Vcore and 1.25625Vcore in less than a minute.

I will lower the multi to x8.

[jimmsch]

OK, running the above settings with two changes.

Vcore now = 1.25625
And the multi is now x8

Small FFT's good for 5 minutes
Large FFT's good for 5 minutes
Blend test good for 5 minutes

Trying 445x9 @ 1.26250Vcore I will go as high as I can on the 9x multi as long as I can keep it under 1.30Vcore stable for 8+ hours of large FFT's

445x9 @ 1.26250 Small FFT crash less than 2 seconds. Trying 1.26875 now.
Same.

Will up Vcore to next notch. 1.2750 Fails in about 1 minute.
Going up one more notch on Vcore

1.28125Vcore also fails small FFT's in about 1 minute

One core failes Large FFT's in 3 minutes.

Next stop 1.2875Vcore
Small FFT fail 2 minutes. Highest core temp 61
Large FFT's did not fail after 6 minutes. Highest core temp 62

Next stop 1.29375 Vcore
Small FFTs fail in 3 minutes.
Large FFTs running 6+ minutes so far.

Large FFTs fail at 8 minutes.

Lowered multi to 8.5...445FSB...1.29375 Vcore...1.90 Vdimm
Small FFTs good for 5+ minutes
Large FFTs good for 5+ minutes
blend test good for 5+ minutes

Tried 450x8.5 = no POST @ 1.29375 Vcore
Upped Vdimm to 2.0v but would not load Windows.

Current settings as listed above with these exceptions.
FSB = 445 x9...Vcore = 1.3000...Vdimm = 1.90
Small FFTs good for 5+ minutes. Core 0 reached 70°C
Large FFTs good for 5+ minutes. Hottest core (core 0) 63°
One core failed blend after 8 minutes

Raised Vdimm to 1.95 and running blend test again
Blend so far so good for 15+ minutes
Large FFTs so far so good for 19+ mnutes

I need to sleep. I will let the large FFTs run overnight and get RIGHT back to this thread in the morning.

Thanks for all your help Jason.

[jason4207]

Like I said, try to focus on 1 thing at a time. You're trying to do it all at once.

Lower that multi to 6 for now and leave vcore at 1.25v or close.

Work on getting 445 stable for now w/ large-fft. 485 will be a lot harder. You'll probably hit a large-fft FSB wall around 460-463 even w/ everything tweaked.

Robust Graphics Booster = FAST
CPU Clock Ratio = x6
Fine CPU Clock Ratio = 0.0 (greyed out...is available at lower multis)
CPU Frequency = (445 x6)
CPU Host Clock Control = Enabled
CPU Host Frequency = 445
PCI-E Freq(Mhz) = 100
CIA 2 = Disabled
Performance Enhance = Standard
System Memory Multiplier = 2.00D (1:1)
Memory Frequency = 890
DRAM Timing Selectable (SPD) = Manual = 5-5-5-15

Advanced Timing Control
Act to Act Delay = Auto (3)
Rank Read to Write Delay = Auto (3)
Write to Precharge Delay = Auto (6)
Refresh to Act Delay = Manual (42)
Read to Precharge Delay = Auto (3)
Static tRead Value = Auto (8)
Static tRead Phase Adjust = Auto (0)
Command Rate (CMD) = Auto (2)

Clock Driving and Skew Control
CPU\PCIEX Clock Driving Control = 800mV This never did anything for me above 700mv except on 65nm CPUs, but you should always experiment to see what your chip likes.
CPU Clock Skew Control = Normal

System Voltage Control
Manual
DDR2 Overvoltage Control = +0.10 Since your RAM can do 1000 5-5-5-15 2.1v let's keep it at at least +0.20 for 890...focus on just the FSB for now...you can always try to reduce this later if you want
PCI-E Overvoltage Control = Normal
FSB Overvoltage Control = Normal Experiment w/ this. See if 1 notch up hurts or helps...it should help...might need more than 1 notch, but be meticulous...BTW, this is the same as VTT
(G)MCH Overvoltage Control = Normal (it seems whenever I change this it results in no POST) If it doesn't help now that is fine. But do try it again later after you've made progress w/ other settings and you hit another wall...this is the same as vNB
MCH Reference Voltage Control = Normal This is NB GTL. Don't be afraid to play; it's not an applied voltage, just how the signal is read...and you may have to come back to this setting as I described earlier
DDR Reference Voltage Control = Normal Is there an auto setting? When experimenting w/ GTLs you should count auto as just another choice to experiment w/...this setting isn't that important, though in my experience.
DDR Termination Voltage Control = Normal
CPU GTLREF1 Voltage Ratio = Auto Why is 1 on auto and 2 on normal? GTLs can be very frustrating to experiment w/, but for 445FSB you should be able to put both at auto or both at normal.
CPU GTLREF2 Voltage Ratio = Normal ^^
Loadline Calibration = Enable
CPU Voltage (Vcore) = 1.25


Don't worry about the temps too much yet as everything isn't dialed in yet. And 70 really isn't that bad. What kind of cooler do you have, and how is your case cooling? What is your thermostat set to if you are running A/C?

[jimmsch]

OK, thanks Jason. I did actually see your post before I went to bed last night and will follow your suggestions. This is what I ran.

Robust Graphics Booster = FAST
CPU Clock Ratio = x6
Fine CPU Clock Ratio = 0.0 (greyed out...is available at lower multis)
CPU Frequency = (445 x6)
CPU Host Clock Control = Enabled
CPU Host Frequency = 445
PCI-E Freq(Mhz) = 100
CIA 2 = Disabled
Performance Enhance = Standard
System Memory Multiplier = 2.00D (1:1)
Memory Frequency = 890
DRAM Timing Selectable (SPD) = Manual = 5-5-5-15

Advanced Timing Control
Act to Act Delay = Auto (3)
Rank Read to Write Delay = Auto (3)
Write to Precharge Delay = Auto (6)
Refresh to Act Delay = Manual (42)
Read to Precharge Delay = Auto (3)
Static tRead Value = Auto (8)
Static tRead Phase Adjust = Auto (0)
Command Rate (CMD) = Auto (2)

Clock Driving and Skew Control
CPU\PCIEX Clock Driving Control = 800mV This is what the mobo defaults to, I just left it there
CPU Clock Skew Control = Normal

System Voltage Control
Manual
DDR2 Overvoltage Control = +0.20
PCI-E Overvoltage Control = Normal Options are +.05 - +.35 in .05 increments
FSB Overvoltage Control = Normal Options are +.05 - +.35 in .05 increments, I assume I will play with this once I start hitting stability issues?
(G)MCH Overvoltage Control = Normal This goes from +.025 to + .375 in .025 increments, I don't know what the starting point is, like with the DDR2 voltage I know it starts at 1.8 and goes up +.05, +10 etc.
MCH Reference Voltage Control = Normal This has two options + or - 10%
DDR Reference Voltage Control = Normal This has two options + or - 10%
DDR Termination Voltage Control = Normal This has two options + or - 50mV
CPU GTLREF1 Voltage Ratio = Options are +3.00%, -3.00%, -6.00%, Auto and Normal
CPU GTLREF2 Voltage Ratio = Normal This did not have an auto setting, so I set them both to Normal...Options are -3.00%, -6.00%, -9.00% and Normal
Loadline Calibration = Enable
CPU Voltage (Vcore) = 1.2375 stock




So I am stable for 5+ hours with large FFTs at 445 x6 using 1.2375 Vcore and 2.0 Vdimm

Where do I go from here?

Thank you for going into such great detail for me. From now on I will do what you suggest.

I tried 450 x6 with these settings and got no POST. Which setting do I try to tweak first?

OK I tried many different settings to POST at 450 x6, but never a combination of any 2 at the same time.
I tried:
raising DDR2 Overvoltage to +.30 no POST
raise FSB Overvolt to +.05 no POST
raising MCH Overvoltage to +.125 on/off cycle - manual clear CMOS
raising MCH ref +10% no POST
raise DDR2 ref no POST
raise Vcore to 1.2500 no POST

I also tried these settings simultaneously.
DDR2 Overvolt +0.30
MCH REF +10%
FSB Overvolt +0.10
Vcore 1.2500
Result in no POST

One thing in common on all no POSTS is it fails o read RAM. It reads CPU speed and multi, but never reads the RAM. I absolutely will not put more than 2.1v into the RAM, and will not run 24/7 with any more than 2.05 Vdimm

If you have ideas on how to boot at 450 x6 FSB, I will try, but 445 FSB is all I need. My goal is 4.0 GHz, which I get if I can get 445 stable on 9 multi.
So, if possible, I would like to have your help in running up the multis using 445 FSB, which seems to be a stable FSB.

Thanks -Jim

Originally Posted by Jason4207

Don't worry about the temps too much yet as everything isn't dialed in yet. And 70 really isn't that bad. What kind of cooler do you have, and how is your case cooling? What is your thermostat set to if you are running A/C?

It is summer, the room is warm, I have no A/C. I imagine the room exceeds 80°F during the day. I have a simple oscillating fan on a pedestal moving the air around the room for me.

P.S. My cooling is Zalman 9700 fan speed set in BIOS for "voltage" It stays quiet that way.

Case air flow is rather good. I have 2 side intakes (1 x120 and 1 x80) the 80 blows directly on the RAM and the 120 blows on the vid card and the NB.
I have 1 x120 front intake.
I have 2 x120 exhausts (1 in top and 1 in rear).
All fans are wire volt modded to 7v (I think, might be 5v, but I think it was 7v) for quietness. Positive pressure inside case keeps dust out.

[jason4207]

Originally Posted by jimmsch

OK, thanks Jason. I did actually see your post before I went to bed last night and will follow your suggestions. This is what I ran.

Robust Graphics Booster = FAST
CPU Clock Ratio = x6
Fine CPU Clock Ratio = 0.0 (greyed out...is available at lower multis)
CPU Frequency = (445 x6)
CPU Host Clock Control = Enabled
CPU Host Frequency = 445
PCI-E Freq(Mhz) = 100
CIA 2 = Disabled
Performance Enhance = Standard
System Memory Multiplier = 2.00D (1:1)
Memory Frequency = 890
DRAM Timing Selectable (SPD) = Manual = 5-5-5-15

Advanced Timing Control
Act to Act Delay = Auto (3)
Rank Read to Write Delay = Auto (3)
Write to Precharge Delay = Auto (6)
Refresh to Act Delay = Manual (42)
Read to Precharge Delay = Auto (3)
Static tRead Value = Auto (8)
Static tRead Phase Adjust = Auto (0)
Command Rate (CMD) = Auto (2)

Clock Driving and Skew Control
CPU\PCIEX Clock Driving Control = 800mV This is what the mobo defaults to, I just left it there That's fine
CPU Clock Skew Control = Normal

System Voltage Control
Manual
DDR2 Overvoltage Control = +0.20
PCI-E Overvoltage Control = Normal Options are +.05 - +.35 in .05 increments
FSB Overvoltage Control = Normal Options are +.05 - +.35 in .05 increments, I assume I will play with this once I start hitting stability issues? Yes, if you are hitting a FSB wall this might help, but it's complicated due to the relationship w/ MCH Ref
(G)MCH Overvoltage Control = Normal This goes from +.025 to + .375 in .025 increments, I don't know what the starting point is, like with the DDR2 voltage I know it starts at 1.8 and goes up +.05, +10 etc. Start on X38 I believe is 1.25v. You can go all the way to 1.45v pretty safely, but you shouldn't need to go that high. 45nm CPUs just don't respond the same as 65nm.
MCH Reference Voltage Control = Normal This has two options + or - 10% Make sure to try all options here (+ and -). The best GTL could be higher or lower...remember it is not an applied voltage, but just how the voltage signal is interpreted. Also try auto if it's an option.
DDR Reference Voltage Control = Normal This has two options + or - 10%
DDR Termination Voltage Control = Normal This has two options + or - 50mV
CPU GTLREF1 Voltage Ratio = Options are +3.00%, -3.00%, -6.00%, Auto and Normal
CPU GTLREF2 Voltage Ratio = Normal This did not have an auto setting, so I set them both to Normal...Options are -3.00%, -6.00%, -9.00% and Normal
Loadline Calibration = Enable
CPU Voltage (Vcore) = 1.2375 stock

You don't have a large number of options for some of the other settings, so you could try to get the FSB right on the edge of stability again (try something b/n 445 and 450 so that you can post, but large-fft fails fairly quickly). And just try each of them in turn to see if you can see any improvements. Remember to try both + and - on all the Refs (CPU, DDR, and MCH). You could also try both of the DDR Term settings.



So I am stable for 5+ hours with large FFTs at 445 x6 using 1.2375 Vcore and 2.0 Vdimm

Where do I go from here?

You could just call the FSB stable, but I would run the test a little longer just to be sure. If you watch the FFT size it will eventually reset. I like it to go through the whole range, b/c if there is a problem when it runs this long it is usually a particular FFT size that is the culprit. And if it keeps failing at the same FFT size you can run a custom FFT size and focus on that particular FFT size instead of waiting 5+ hrs to get back where you were.

I would also want to go a little higher on the FSB, find the best option for every setting, and then back down later. I don't like my FSB to be only a few MHz from failure b/c it can sometimes cause random BSOD later

But if you want to move along you could just leave it, crank that multi up 9, and see what happens. Hopefully it'll boot just fine, but you should only need to play w/ vcore (possibly CPU GTLs too if you want to get vcore as low as possible) to get it small-fft stable.

Thank you for going into such great detail for me. From now on I will do what you suggest.

I tried 450 x6 with these settings and got no POST. Which setting do I try to tweak first?

See above, but i think you should try MCH Ref at -10% first.

OK I tried many different settings to POST at 450 x6, but never a combination of any 2 at the same time.
I tried:
raising DDR2 Overvoltage to +.30 no POST
raise FSB Overvolt to +.05 no POST
raising MCH Overvoltage to +.125 on/off cycle - manual clear CMOS
raising MCH ref +10% no POST
raise DDR2 ref no POST
raise Vcore to 1.2500 no POST

I also tried these settings simultaneously.
DDR2 Overvolt +0.30
MCH REF +10%
FSB Overvolt +0.10
Vcore 1.2500
Result in no POST

One thing in common on all no POSTS is it fails o read RAM. It reads CPU speed and multi, but never reads the RAM. I absolutely will not put more than 2.1v into the RAM, and will not run 24/7 with any more than 2.05 Vdimm

Keep in mind that the NB and RAM are intimately intertwined. It is hard to distinguish the difference as to which may be at the heart of your problem. That's why it's a good idea to keep the RAM at known good settings (or looser & slower) until you get the NB stable. Then you can be more certain where the issue lies.

If you have ideas on how to boot at 450 x6 FSB, I will try, but 445 FSB is all I need. My goal is 4.0 GHz, which I get if I can get 445 stable on 9 multi.
So, if possible, I would like to have your help in running up the multis using 445 FSB, which seems to be a stable FSB.

Go for 9x! Let us know if it runs, and start using small-fft to focus the stress more on the CPU, and use vcore (and perhaps CPU GTLs) to try and get it stable.

Thanks -Jim



It is summer, the room is warm, I have no A/C. I imagine the room exceeds 80°F during the day. I have a simple oscillating fan on a pedestal moving the air around the room for me.

P.S. My cooling is Zalman 9700 fan speed set in BIOS for "voltage" It stays quiet that way.

Does it get loud at 100%? If not, run it at 100% 24/7. I am under the impression that the Zalmans are pretty quiet even on 100%, but I've never had one. With your high ambients it might help as well.

Case air flow is rather good. I have 2 side intakes (1 x120 and 1 x80) the 80 blows directly on the RAM and the 120 blows on the vid card and the NB.
I have 1 x120 front intake.
I have 2 x120 exhausts (1 in top and 1 in rear).
All fans are wire volt modded to 7v (I think, might be 5v, but I think it was 7v) for quietness. Positive pressure inside case keeps dust out.

Good job on the + pressure! Keeps the dust out of all the cracks. Run them as high as you can stand them. If they are on 5v, try 7v and see if the noise is tolerable.

[jimmsch]

OK, I did a little better tha 30 minutes at 445 x8.5...1.2375Vcore...2.0Vdimm on P95 blend with no errors.


And I did about 90 minutes at the same settings on P95 large FFT


I am now trying to get P95 to run ANY test for morethan a minute on 445x9.
I have been messing with the MCH ref and the FSB ref, but I will set them to auto now and go with Vcore and maybe GTL as Jason suggests.

[jason4207]

It seems to be pretty stable at 445FSB w/ your previous MCH Ref and FSB ref, so I'd leave those setting alone.

Just keep it at 445x9 and try and see if more vcore helps at all. CPU-Z is showing only 1.2v for vcore. The BIOS vcore setting is just a label, so always go by CPU-Z and adjust the vcore up or down in the BIOS as needed. You're going to need to see b/n 1.25-1.3v in CPU-Z to be stable at 4GHz. It's best to use small-fft now.

Later you can double-check everything by running all 3 tests for longer.