OC E6600/680i produces terrible results. Help?

[cronjob]

I've spent the last two days attempting to overclock my system even a bit and I'm coming up with unreasonable results. Could someone familiar with my setup offer some guidance? I haven't overclocked in a number of years and only ever on AMD, so I am a little bit out of my comfort zone and concerned if I'm doing something stupid.

Problem: Even with 40% CPU overclock, 3dmark06 scores only increase 10% at the cost of at least a 20c increase!

* Yes, I know that 3dmark06 is overrated, but it's what everyone uses so it gives me something to compare my overclocking improvements to other people who, with a similar setup but only an E6300, get 16,000+ scores.


Hardware
C2D E6600
eVGA 680i
2x1gb Crucial Ballistix PC6400 4-4-4-12
eVGA 8800 GTX
Tuniq 120 HSF
P&C Power & Cooling 750w Silencer
1x120 intake fan (front of case)
1x120 exhaust fan (rear)
3x140 intake fans (panel)


All of the following configurations included turning off CPU Thermal Control, C1E Enhanced Halt State, Intel Speedstep, Spread Spectrums, SLI-Ready Memory, Linkboost, Smartfan and NVidia GPU Ex. Additionally, they are all done using the Linked and Synced 1:1 method. All heat/stress testing done with Prime95. All temperature readings done with Core Temp.

Default Configuration: FSB 800, 2400mhz
Stock settings, voltages autoset, timings autoset.

4-4-4-14 1T

FSB: 800 (2400mhz)
Core: 1.2v
FSB: 1.2v
RAM: 1.85v (Crucial says the RAM should be 2.2v)
SPP: 1.2v
MCP: 1.5v
HT/SPP: 1.2v


Idle: 38c
Stress: 45c
3dMark06: 9,796



Default Configuration: FSB 1200, 2700mhz
Stock settings, voltages autoset, timings manually set.

4-4-4-14 1T

FSB: 1200 (2700mhz)
Core: 1.2v
FSB: 1.2v
RAM: 1.85v
SPP: 1.2v
MCP: 1.5v
HT/SPP: 1.2v


Idle: 39c
Stress: 51c
3dMark06: 10,241



Default Configuration: FSB 1300, 2925mhz
Stock settings, voltages autoset, timings auto set.

4-5-5-15 1T

FSB: 1200 (2700mhz)
Core: 1.2v
FSB: 1.2v
RAM: 1.85v
SPP: 1.2v
MCP: 1.5v
HT/SPP: 1.2v


Idle: 39c
Stress: 51c
3dMark06: 10,464



Default Configuration: FSB 1400, 3150mhz
Stock settings, voltages autoset, timings manually and auto set.

4-4-4-12 2T (manual)
5-5-5-15-1T (manual)
5-5-5-17 2T (autoset)

FSB: 1200 (2700mhz)
Core: 1.2v
FSB: 1.2v
RAM: 1.85v
SPP: 1.2v
MCP: 1.5v
HT/SPP: 1.2v


Idle: 45c
Stress: 70c
3dMark06: 10,635



Default Configuration: FSB 1500, 3375mhz
Stock settings, voltages autoset and manually set, timings manually set.

4-4-4-12 2T / 4-4-4-12 1T
3-3-3-9 2T / 3-3-3-9 1T

FSB: 1200 (2700mhz)
Core: 1.2v
FSB: 1.5v (manually set - default 1.2v)
RAM: 2.2v (manually set - default 1.85v)
SPP: 1.2v
MCP: 1.5v
HT/SPP: 1.2v

Idle: 48c
Stress: 71c
3dMark06: 10,782 to 10822

 

[GrimReaperGuy]

I don't see too much of a problem, 3dMark06 isn't particularly CPU-bound. If you were expecting linear scaling with your CPU overclocking, you were mistaken. If you're aiming to increase your 3dMark, overclock the graphics card and memory some (or more).
I am, however, quite astounded that you pushed your E6600 to 3.375GHz on 1.2vCore.

 

[jsc]

I have a similar system - GTS insead of GTX and PC8000 Crucial instead of PC6400.

For some reason, 3DMark06 does not run. Computer sits there, and when I pop up Task Manager, I get a "Not responding".

Using 3DMark05, I got a fairly linear score increase from 2.4 GHz to 2.8 Ghz. Then the scores seem to level off past 3.0 GHz to 3.6 Ghz..

My conclusion is that there is nothing wrong here. I think that the C2D processors are fast, powerful, and efficient enough to keep the 8800's fed.

I am changing my signature to "e6600 at 3.4 Ghz" and mem timing to 3-3-3-7. System runs at 3.6 Ghz and core temps are OK (8 hours of Ortho) but I don't like running with core voltage at 1.5 volts. Besides, the more aggressive memory timing gives me about the same throughput as judged by the cpu scores in 3DMark.

I am not an expert in C2D systems, but it seems to me that your core temps are higher than they should be with a Tuniq 120.

As far as the memory is concerned, great! It drops the thermal load on your system.

john

 

[cronjob]

But if I'm not mistaken, the 8800 GTX is very limited by current CPUs which would suggest that increasing performance of the CPU would open up some bandwidth in which the GPU can play, right?

As for voltages.. I'd love to understand how the 680i BIOS figures it out, because the values it uses to autoset are complete crap.

 

[jsc]

Cronjob,

I wouldn't say "crap". They are a little stinky though. On my board, I need a BIOS set of about 1.53 v to give me an indicated 1.5 volts in System Monitor. Another reason I dropped the speed. I didn't know which to trust.

john

 

[cronjob]

I'm a little confused now.

I manually set my timings to 4-4-4-12 1T

CPU-Z reports that is what I have it set to.

Everest says:

DIMM1: Crucial Tech. 1 GB DDR2-800 DDR2 SDRAM (5-5-5-18 @ 400 MHz) (4-5-5-15 @ 333 MHz)

DIMM2: Crucial Tech. 1 GB DDR2-800 DDR2 SDRAM (5-5-5-18 @ 400 MHz) (4-5-5-15 @ 333 MHz)

Also, I can't figure out what the real vcore voltage is supposed to be. CoreTemp has a 1.2375 VID while CPU-Z says 1.125. More, the BIOS is completely whack and even if I have it set to 1.4v it shows 1.36 as the "current" voltage when I restart.

 

[techgeek]

If you are using the free version of Everest, you can't go by what it says. I had the same issue with my motherboard. The last free version is something like 2.2 (very old) and doesn't support new hardware. I purchased the newest version and it fixed the erroneous readings (core voltage being one of them). If you want to see RAM timings and core voltage get CPU-Z (gizoogle it), it will report them correctly.

EDIT: Sorry I see that you mentioned using CPU-Z so disregard that part. Just know that if you are using the free version of Everest, much of the information it gives will be incorrect with your set up.

 

[cronjob]

I'm actually running Everest 3.01. Perhaps I am just reading it wrong.

 

[techgeek]

I'm betting if you upgraded that to 3.5, you would straighten out at least the Core voltage. It would also display both core temperatures (like CoreTemp and TAT).

 

[cronjob]

Oh, I didn't even realize there was a newer version out now. Thanks for point it out.

 

[jsc]

techgeek,

OK. I just went out and downloaded CPU-Z. Now I am REALLY confused. CPU-Z seems to verify what I think is going on except for ... (drumroll)
CPU core voltage. BIOS core set voltage indicates 1.50 volts. System Monitor indicates 1.45 volts. CPU-Z shows 1.213 v. ????

You are saying that CPU-Z is the most accurate?

If true, that means two things. First, I owe Cronjob an apology. It IS a piece of crap. Second, I am running at 3.4 Ghz on 1.2 v. 8O

BTW, what are you doing in Yemen?

john

 

[cronjob]

I had to set my BIOS core voltage to just under 1.37 to get it to show up in the "Current" column of the BIOS as 2.35ish.

Then, booting into Windows, Core Temp Beta 0.94 shows the VID should be 1.2375v.

CPUZ says, at the moment, 1.240 (I'm going to guess it's rounding-up?).

Everest Ultimate Edition 2006 (3.01) currently is showing the core voltage as 1.238.

So now that I've changed the core voltage in the bios a little more, it seems to be getting more accurate. I"m not sure why there was previously such a big difference. Still a bit frustrating that what you set in the BIOS and what the BIOS reports is so different.

I'm a little worried that what I'm setting might be what it's really at and what it's REPORTING might be inaccurate. So I might be cranking it up to well over the VID on the left column of the BIOS just to make it appear properly on the RIGHT side of the BIOS... and perhaps because of the discrepency, I might be over-volting it a bit? Does anyone know if the reported (in the bios) value is actually the legit value?

 

[techgeek]

Is that reading in windows of 1.2V under load, do you have something like Folding running putting your system under load. Even that wouldn't totally explain it.

There is something known as Vdroop, and you may have heard of Vdroop mods. Basically it is built into Intel's (and possibly AMD's) VRM guidelines that under load Vcore should load down by some percentage. This is supposed to help with thermals (ie under load Vcore lowers slightly, lowering overall power dissipation). Unfortunately because of Vdroop, it makes it more difficult to overclock. As you increase Vcore, to get your speed up, Vcore droops more under load. This results in bigger swings from loaded to unloaded. When sudden changes in CPU load occur in these situation, your system could become unstable (spontaneous reboots, lockups). The Vdroop mods undo or decrease the amount Vcore is allowed to drop under load, resulting in a smaller swing in Vcore. Thus making it more stable in an overclocked condition.

Like I said I don't think Vdroop could account for all the difference from what you are seeing in your BIOS monitoring, and in Windows.

As for what you set Vcore to in the BIOS and what it reads in Windows, all these values have to be taken with a grain of salt. The only way to accurately measure Vcore is to get a voltage meter and measure the output of the VRM. Of course this isn't something most users should attempt unless they are very familiar with electronics. There are risks to sticking meter leads into your motherboard, namely grounding out something and destroying your motherboard or more. The problem lies in the fact that software is calculating the values for all the voltages, fan speeds, temperature coming off the SM Bus. These values are quantized values (converted from analog to digital) in binary form. These values are run through a formula which calculates the values in units we can understand. So the accuracy of these values are dependant on the accuracy of the formulas. Often the values read in the BIOS is inaccurate (I know you'd think it would be the other way around). Same goes for the values you set for Vcore, they won't necessarily read what you set them to, as is the case for you. That's not saying that CPU-Z is correct either. Sometimes it takes getting two corresponding values from different software to pick out the most likely value. Again though the only true way to know is to actually take a physical measurement. As long as everything is stable, it's not too big a deal. One thing to note is that the software you are using to measure your values has to be able to profile your motherboard (and more importantly the chip used to collect the readings). Each motherboard is different. The newer your motherboard, the less likely you'll get accurate calculations for your values within the software. That's why I suggested to cronjob to get the newest version of Everest. The newest versions are likely to have support for the latest hardware.

As for why I'm in Yemen, I am an Electronics Engineer working in the oil industry. I work here, live in Canada.

EDIT: One thing to note about the values read in the BIOS. Quite often these values are inaccurate and are fixed during subsequent BIOS updates. I have found in particular ASUS boards read temperatures way lower than third party software does in Windows. So low in fact that even without the software discrepancy, you still know there off because they are unbelievably low.

 

[techgeek]

Hey cronjob, some of my post to jsc applies to you. As for your reading, you have three programs pointing to ~1.23, so I would trust that. Again like I mentioned to jsc short of actually measuring, you don't know exactly what Vcore is.

I gave your original post a closer look, and your last overclock @ 3375, your temps are a little high there. If you can't improve airflow/cooling, I would back off to somewhere in between 325 and 350 (closer to 325). I wouldn't run at 70C 24/7, it maybe stable, but that temp long term is going to shorten the life of that CPU. You mention that you are using Prime95 to do your stress testing, are you running two instances of it to utilize both cores? If not that temperature is even more scary. I found a good program to find the thermal envelope of a system is TAT. You can load each core to 100%. This program heats up the CPU better than Prime, Othoros (which is a fancy front end for Prime that makes running it on two cores easier), OCCT, Super PI. I don't know if there's any checking for errors as in Prime95 (at least it hasn't errored while I ran it), but it maybe something a person wants to try after he has established stability with Prime95. It definitely gives you worst case scenario in terms of temperature.

If your are concerned about where you have your Vcore set at, lower it a notch and test for stability, if it works lower it another notch and so on until it becomes unstable, then bump it up one or two. You seem to be at your thermal limits as it is, so lowering Vcore could drop that temperature down to managable levels.

 

[jsc]

More testing.

I have been using Orthos to load both cores to check for stability and thermal loads. If I start CPU-Z while idling, it shows 1.213 volts. Once Orthos starts, CPU-Z indicates 1.44 v (very close to what the BIOS indicates). Within 1 second of stopping Orthos, it drops back to 1.213 volts. If I disable 1 core in the BIOS, the cycle is 1.213 volts to 1.448 volts back to 1.213 volts. That does explain why some of us are seeing greater than 3 GHz overclocks on an indicated 1.2 volts.

Puzzlement. Evidently, the BIOS is showing what the loaded voltage will be while in an unloaded (pre-boot) state.

I wish I knew what the voltage regulator module on this board is doing. There has to be test points for factory testing somewhere.

My core temps are staying below 50 degrees C under load.

I am a retired U.S. Army electronic technician mostly doing training for the Saudis.

 

[techgeek]

That seems to me like SpeedStep or some other power save function is in operation. SpeedStep reduces clockspeed and/or Vcore during idle times. Cronjob could you try the same thing? Cronjob mentioned in his original post that he had SpeedStep and other power save features turned off, but maybe it isn't being totally disabled. This could be the only reason I could think of for why you are seeing Vcore increase under load. It also would explain why your C2D is able to operate at the clockspeed you have it. I am not familiar with your boards so I couldn't say for sure. It seems like you are on to something though. SpeedStep usually reduces the ability to overclock into the high end.

As for why it shows close to what you set it to in the BIOS is that when you are in the BIOS you are in 16bit Real-Mode. In 16bit Real-Mode there aren't any power save features so you run at whatever Vcore and clockspeed are set in the BIOS. Once you switch to 32bit Protected-Mode, all power save / advanced features are available. Since most people run Windows XP, SpeedStep and Cool n Quiet are built into the operating system (unlike Win 2000 where you had to install the utility). So if they aren't disabled in the BIOS, they are automatically used.

Now this could explain why Cronjob is seeing the exact same thing as you (you haven't explicitly said that EIST is disabled in your BIOS so I can't lump you in with this assumption), maybe the BIOS isn't totally disabling EIST, leaving the voltage to be manipulated within Windows (via SpeedStep). If this is the case the motherboard manufacturer needs to update their BIOS to totally disable it when you select to disable it. These are still relatively new boards so this doesn't seem far fetched.

 

[jsc]

I did some more testing. I double checked. Both BIOS and XP report that SpeedStep is off. The 1.2v - 1.44v - 1.2 volt cycle seems to happen only at some voltage settings. At other settings, CPU-Z reports about a .015 volt drop under load compared to idling conditions.

At any rate, I am glad I resisted the temptation to look at the 1.2 volt reading and assume that I had another .3 volt to play with.

john

 

[Latta]

According to the evga website the most up to date BIOS version is said to correct incorrect voltage readings in the BIOS.

--------------------------------------------------------------
BIOS Version : P24 - EVGA Recommended Download
BIOS Date : 1/12/2007
BIOS File : NF68_P24.bin (filesize N/A)
Notes : The following was updated in release P24:

Vista WHQL Certified
USB Floppy improvements for RAID installs
Correct CPU temp and voltage system monitor displays in BIOS
Correct default CPU multiplier setting
Fix "Code 50" hang received under certain VGA configurations
Also includes these updates from bios P20, P21 and P23:

Fix for SATA disk drives
Improvements to memory stability and overclockability
Disabled spread spectrum tables for improved overclockability
Improve POST screen CPU speed reporting
Improves X-Fi and internal audio functionality
WHQL Certified BIOS
Improvements to overclocking
Enhancements for Quad-Core CPUs
Adds ability to enable splash screen

 

[cronjob]

Well, then I'd say they need to give it another attempt in P25.

 

[Latta]

Apparently a lot of people according to the forum are having issues installing the P24 update correctly.

It appears as though they run the exe and think its been updated but in fact theres a bin file you have to execute. From what I gather. I don't have the board so I can't direct you any further but I would verify you have P24 installed.

 

[cronjob]

You have to select the downloaded *.bin from inside the Update utility to flash it, causing a reboot. You can tell if you have it because it will mention something about the version including "P24" (or whatever version you have) during POST.

I'm pretty happy with the board except I'm not having great overclocking experiences at the moment and I really wanted to get a good Creative Labs card (which obviously is pointless with this chipset). Hopefully they'll be able to fix everything via software rather than hardware.

 

[Habber]

Cronjob when you say you dont get good overclocking results what kinda results are we talking about?

My board works fine, and it even overclocks without me even having to adjust the voltage. first time i went straight for 3.2mhz (i have an E6600) only adjusted the fsb and unlinked the ram modules and it ran fine. 5 mins later i went to bios and set it to 3.6mhz without a problem. has been running fine for over 2 weeks now. have idle temps around 32 and load around 55 with an Noctua NH-U12F cooler. cant complain about this at all. performs around 11500 in 3dmark06 with and 8800gtx and 2 gigs of corsair.