C-State and OC Stability

[QuickShow]

Hi there.

I've been trying to reach stability with highest multiplier I can reach. The multiplier is set @ 26 with BCLK @ 170, giving a CPU clock = 4.42 GHz. vCore = 1.325V, vQPI = 1.275V, LLC = 100%

With C0, C1, and SpeedStep disabled the system passes LinX 20 runs, renders Cinebench 10 runs, and finishes 3DMark 11 benching. But when these power-saving functions are enabled the system even freezes during 3DMark 11 benching.

So my question is, is C-State or/and SpeedStep chewing up the stability of the system? If yes should I disable them or should I bump up the voltages up by a notch to stabilize the system? Both are going to generate more heat for sure.

 

[Idonno]

Yes all that power saving stuff will mess with your OC stability as well as "CPU Spread Spectrum" and PCI Spread Spectrum

24GB ram filling all 6 dimms highly stresses my CPU's IMC's and is probably my biggest challenge when OCing. I run my QPI/DRAM core volts @ 1.39375v for a 24/7 completely stable 4.3GHz OC. With 4GB of ram you should be able to keep yours lower.
Also with 4GB instead of 24GB your ram should be much easier to OC

I have found I have better results with stability running lower latency and voltage ram settings of 8-8-8-20 CR1 @ 1.5635v @ 1605MHz, instead of the recommended 9-9-9-24 @ 1.65v @ 1600MHz with a 160 BCLK and 27 CPU multi than I do trying to increase the rams speed.

Raising the CPU PLL to 1.84175v and the IOH voltage (NB): 1.20575v helped to stabilize these settings as well. At the stock PLL of 1.81v @ 160 BCLK I had rare but occasional issues. Now at 1.84175v PLL those issues are gone. Both these setting are intertwined a bit but the higher IOH was mostly for my Ram so you might not need it and the CPU PLL is more closely related to BCLK which might be of some help to you.

Settings for my 24/7 4.3GHz OC:

CPU Ratio: 27
BCLK Freq: 160
DRAM Freq: 1605Mhz
RAM Timings: 8-8-8-20 / CR1
CPU Voltage: 1.36875
QPI/DRAM core volts: 1.39375
DRAM Bus Voltage: 1.5635
CPU PLL: 1.84175
IOH Voltage (NB): 1.20575
IOH PCIE Voltage (NB): Auto (1.51)
ICH Voltage (SB): 1.113
ICH PCIE Voltage (SB): Auto (1.51)

More Bios settings:

Sync Mode: Enabled
AI Overclock Tuner: Manual
OC From CPU Level Up: Auto
2D Performance Booster: Disabled
CPU Turbo Power Limit: Disabled
PCIE Frequency: 100
Memory Recheck: Enabled
Memory Configuration Protect: Disabled
CPU Differential Amplitude: Auto
CPU Clock Skew: Auto
IOH Clock Skew: Auto
Digi+ Pwr Mode: T-Balanced
PWM Volt. Control: Auto
Load-Line Calibration: Full Calibration
CPU Voltage OCP: Enabled
CPU PWM Frequency: Auto
Extreme OV: Disabled
Extreme OC: Auto
CPU Spread Spectrum: Disabled
PCI Spread Spectrum: Disabled

Advanced CPU Settings:

C1E Suppport: Disabled
Hardware Prefetcher: Enabled
Adjacent Cache Line Prefetch: Enabled
Intel® Virtualization Tech: Disabled
CPU TM Function: Disabled
Execute Disable Bit: Disabled
Intel HT Techology: Enabled
Active Processor Cores: All
A20M: Disabled
Intel SpeedStep Tech: Disabled
Intel C-STATE Tech: Disabled

I have recently increased my 24/7 OC to 4.5GHz
These are the only settings I changed:

CPU Ratio: 28
CPU Voltage: 1.3875

I have been running this for 3 days now w/no issues. My PC does fine with "Intel burn Test" @ "very high" settings 30 passes and "AIDA64-System Stability Test" (w/all options checked except "stress local disks") over night.

Another issue I had was some instability shortly after my NB/SB temps went over 65c and I had to rearrange my fan configuration to take care of that. Now keeping NB/SB temps 65c or lower my OC seems 100% solid.

Anyway, we have the same MB and almost the same CPU (unless you also have the X version) so I thought this might be helpful.

 

[QuickShow]

Yes all that power saving stuff will mess with your OC stability as well as "CPU Spread Spectrum" and PCI Spread Spectrum

24GB ram filling all 6 dimms highly stresses my CPU's IMC's and is probably my biggest challenge when OCing. I run my QPI/DRAM core volts @ 1.39375v for a 24/7 completely stable 4.3GHz OC. With 4GB of ram you should be able to keep yours lower.
Also with 4GB instead of 24GB your ram should be much easier to OC

I have found I have better results with stability running lower latency and voltage ram settings of 8-8-8-20 CR1 @ 1.5635v @ 1605MHz, instead of the recommended 9-9-9-24 @ 1.65v @ 1600MHz with a 160 BCLK and 27 CPU multi than I do trying to increase the rams speed.

Raising the CPU PLL to 1.84175v and the IOH voltage (NB): 1.20575v helped to stabilize these settings as well. At the stock PLL of 1.81v @ 160 BCLK I had rare but occasional issues. Now at 1.84175v PLL those issues are gone. Both these setting are intertwined a bit but the higher IOH was mostly for my Ram so you might not need it and the CPU PLL is more closely related to BCLK which might be of some help to you.

Settings for my 24/7 4.3GHz OC:

CPU Ratio: 27
BCLK Freq: 160
DRAM Freq: 1605Mhz
RAM Timings: 8-8-8-20 / CR1
CPU Voltage: 1.36875
QPI/DRAM core volts: 1.39375
DRAM Bus Voltage: 1.5635
CPU PLL: 1.84175
IOH Voltage (NB): 1.20575
IOH PCIE Voltage (NB): Auto (1.51)
ICH Voltage (SB): 1.113
ICH PCIE Voltage (SB): Auto (1.51)

More Bios settings:

Sync Mode: Enabled
AI Overclock Tuner: Manual
OC From CPU Level Up: Auto
2D Performance Booster: Disabled
CPU Turbo Power Limit: Disabled
PCIE Frequency: 100
Memory Recheck: Enabled
Memory Configuration Protect: Disabled
CPU Differential Amplitude: Auto
CPU Clock Skew: Auto
IOH Clock Skew: Auto
Digi+ Pwr Mode: T-Balanced
PWM Volt. Control: Auto
Load-Line Calibration: Full Calibration
CPU Voltage OCP: Enabled
CPU PWM Frequency: Auto
Extreme OV: Disabled
Extreme OC: Auto
CPU Spread Spectrum: Disabled
PCI Spread Spectrum: Disabled

Advanced CPU Settings:

C1E Suppport: Disabled
Hardware Prefetcher: Enabled
Adjacent Cache Line Prefetch: Enabled
Intel® Virtualization Tech: Disabled
CPU TM Function: Disabled
Execute Disable Bit: Disabled
Intel HT Techology: Enabled
Active Processor Cores: All
A20M: Disabled
Intel SpeedStep Tech: Disabled
Intel C-STATE Tech: Disabled

I have recently increased my 24/7 OC to 4.5GHz
These are the only settings I changed:

CPU Ratio: 28
CPU Voltage: 1.3875

I have been running this for 3 days now w/no issues. My PC does fine with "Intel burn Test" @ "very high" settings 30 passes and "AIDA64-System Stability Test" (w/all options checked except "stress local disks") over night.

Another issue I had was some instability shortly after my NB/SB temps went over 65c and I had to rearrange my fan configuration to take care of that. Now keeping NB/SB temps 65c or lower my OC seems 100% solid.

Anyway, we have the same MB and almost the same CPU (unless you also have the X version) so I thought this might be helpful.

Thanks for such detailed configuration.

Yours settings are pretty much the same as mine except that I have a lower vCore and vQPI. My 12 GB Vengeance just won't do CAS8 so I left DRAM @ stock 1.5V. The CPU Differential AMP (mine @ 700mV) seems to help with stability but I'm not too sure about it. CPU PLL does not seem to help anything at all (or maybe I'm just unaware of it). With all power-saving options on and vCore @ 1.3375V my system seems to be doing fine.

Have you tried running LinX? What's your cooling configuration? What temp do you get?

 

[Idonno]

your lower vQPI is understandable but your vCore is real good for a stable 4.42 OC. Nice! I've never messed with the CPU Differential AMP. In fact your the first one I've heard say much of anything about it. I just leave mine on auto although I'm sure the settings there for a reason.

As far as the CPU PLL just moving it up and down .05 volts or so, to see what direction to head in might not get you any better (or worse) results. 1.84175v gave me better results than the default voltage and 1.855 was OK too but, anything above or below that yielded nothing different than stock. It was about finding the sweet spot (I think).

Turning power saving options off also helped me. Not so much in stress tests but with them on I would get rare but occasional freezes when the power/ratio changed.

I haven't tried LinX but I think I might. It looks very similar to Intel burn Test. I usually use Intel burn Test @ "standard" setting and only 5 or 10 passes to narrow things down, then 10 more passes when I think I may have something usable. After that I'll move on to AIDA64's benchmarks and compare them with previous benchmarks. I'll keep repeating this process until I think I've adjusted every thing as well as possible for my desired OC. Then It's back to Intel burn Test @ "very high" setting for 20 or 30 passes. If all is well, I'll test with AIDA64-System Stability Test (w/all options checked except "stress local disks") for a few hours then overnight if that's all good I'll use my PC for about 1 week to make sure and only then break out Prime 95.

As far as cooling goes I have the CORSAIR Hydro H70 CPU Cooler w/2-fans in a push pull configuration. I have a total of 12 fans in my PC if you include the 3 fans for my video cards and PSU and as long as my temps stay under 80c (not including the occasional quick spike into the lower 80's) while stress testing I'm good with that.

How about you?

 

[Idonno]

I did a little reading on "CPU Differential Amplitude" and it does seem to be the kind of thing that adding a little could help me with long term stability like those "rare but occasional freezes". So thanks for mentioning it. It does seem to add a little heat though.

I also checked out "CPU Clock Skew" and changing that setting had negative effects so I put it back to auto. Ran 4.5GHz for 4 days until my PC crashed and I decided to back off a little so I could keep my CPU around for a little longer.

So, I lowered my rams timings back to the stock 9-9-9-24 CR2, dropped the BCLK to 157 and was able to get my QPI a little lower. I feel more comfortable with these settings. Here's a screen shot.

 

[QuickShow]

Best answer selected by QuickShow.

 

[QuickShow]

I did a little reading on "CPU Differential Amplitude" and it does seem to be the kind of thing that adding a little could help me with long term stability like those "rare but occasional freezes". So thanks for mentioning it. It does seem to add a little heat though.

I also checked out "CPU Clock Skew" and changing that setting had negative effects so I put it back to auto. Ran 4.5GHz for 4 days until my PC crashed and I decided to back off a little so I could keep my CPU around for a little longer.

So, I lowered my rams timings back to the stock 9-9-9-24 CR2, dropped the BCLK to 157 and was able to get my QPI a little lower. I feel more comfortable with these settings. Here's a screen shot.

http://i831.photobucket.com/albums/zz232/Idonno1/Screenshot.jpg

AMP increases noise threshold so theoretically it should help stabilizing the system. In real use I have not yet seen noticeable temperature increase by setting AMP to 700 mV.

For i7 980/980x, I think there is a sweet point beyond which high voltage (Core and QPI) is required and much more heat is generated. From your chip and mine I would say it's probably somewhere between 4.4 and 4.5 GHz.

The IMC on my chip is not awesome enough to drive my sticks beyond 20000MB/s, which was achieved by my old Xeon W3565.

Also, Intel Burn Test does not like my CPU at all; regardless of the high voltage I put on Core and QPI, it can hardly run 2 passes. (Maximum @ 10000MB)

 

[Idonno]

AMP increases noise threshold so theoretically it should help stabilizing the system. In real use I have not yet seen noticeable temperature increase by setting AMP to 700 mV.)

Yes, your right 700's not bad however I also tried 800mV and the temperature increase was quite noticeable so I put it back on 700mV.

As it turns out I must have exited bios without saving when I changed my ram timings because for that screen shot in my last post my ram timing were actually still 8-8-8-20 CR1 but it stress tested stable for 8hrs.

The IMC on my chip is not awesome enough to drive my sticks beyond 20000MB/s, which was achieved by my old Xeon W3565.

Don't feel bad If you fill all 6 dimms you won't get anywhere close to that. I can only OC my Corsair Dominator's to 1785MHz with stock timings and not with a high OC. Right now they're only at 1575MHz. I was able to lower the timing so I guess that's better than nothing.

Also, Intel Burn Test does not like my CPU at all; regardless of the high voltage I put on Core and QPI, it can hardly run 2 passes. (Maximum @ 10000MB)

Don't feel bad I couldn't pass it at your BCLK either (I tried).

And thanks for choosing me for the best answer but, I'm not sure who helped who. So thanks for pointing me in the direction of CPU Differential Amplitude. I believe that's going to be a big help.

 

[QuickShow]

I am now experimenting a 4.6 GHz overclock. With BCLK @ 200 and multiplier @ 23, it runs @ 1.41875 vCore and 1.325 vQPI without crashing. Memory @ 1600 MHz. Temperate max @ 72 °C on one core during 10 run LinX.

 

[Idonno]

I am now experimenting a 4.6 GHz overclock. With BCLK @ 200 and multiplier @ 23, it runs @ 1.41875 vCore and 1.325 vQPI without crashing. Memory @ 1600 MHz. Temperate max @ 72 °C on one core during 10 run LinX.

Sounds good but I don't think I could ever go that high without getting a better CPU cooler. On any stress test my temps would be through the roof!

 

[QuickShow]

Sounds good but I don't think I could ever go that high without getting a better CPU cooler. On any stress test my temps would be through the roof!

I used to have an H80. It cools down the CPU only when it's not overclocked and the ambient temperature is low.

 

[Idonno]

I used to have an H80. It cools down the CPU only when it's not overclocked and the ambient temperature is low.

Since the ambient temperature is always low (72F max) in my computer room, my H70 does OK for what it is. I'm good to 4.4GHz on a long stress test. Maybe a H100 would help me feel safe at 4.5GHz but, anything more than that I'd probably need to get real water cooling (like you have). Do your NB, SB and mosfets have waterblocks as well?

 

[QuickShow]

Since the ambient temperature is always low (72F max) in my computer room, my H70 does OK for what it is. I'm good to 4.4GHz on a long stress test. Maybe a H100 would help me feel safe at 4.5GHz but, anything more than that I'd probably need to get real water cooling (like you have). Do your NB, SB and mosfets have waterblocks as well?

No my motherboard is not watercooled at all, although I have two fans installed on the NB fin. NB temperature can occasionally goes up to 80°C but that's how it is designed. NB temperature does not seem to affect my OC stability.

 

[Idonno]

No my motherboard is not watercooled at all, although I have two fans installed on the NB fin. NB temperature can occasionally goes up to 80°C but that's how it is designed. NB temperature does not seem to affect my OC stability.

My stability starts getting a little erratic after 65c on northbridge. Of course that is stress testing with IntelBurnTest which is always the first stress test I use.

My board also has another issue. All my PCI busses are showing errors "I/O Access Memory Access Bus Master System Errors" due to bios assigning overlapping memory addresses. I just got an RMA# from Asus today. This PC still preforms amazingly well considering but, I've been putting it off for a while and It needs to be fixed.

 

[QuickShow]

My stability starts getting a little erratic after 65c on northbridge. Of course that is stress testing with IntelBurnTest which is always the first stress test I use.

My board also has another issue. All my PCI busses are showing errors "I/O Access Memory Access Bus Master System Errors" due to bios assigning overlapping memory addresses. I just got an RMA# from Asus today. This PC still preforms amazingly well considering but, I've been putting it off for a while and It needs to be fixed.

X58 NB Chipset doesn't really cool down, even with massive waterblock installed. Besides, NB should not affect core stability since IOH only has control over PCIe. Maintaining the NB under 65°C is unrealistic for my system. However, it is suggested by many people that increasing IOH voltage by a few notches helps to stabilize the system (especially with multiple graphics card installed). Since I have a triple SLI my IOH is set @ 1.2V instead of auto.

I/O error rarely happens but if it does it is likely due to hardware deficit.

My case has very bad ventilation; heat is trapped inside and cannot escape from the top. Therefore with an ambient temperature of 30°C the NB will rise to 65°C @ idle and 80°C @ burning.

 

[Idonno]

My IOH is set @ 1.2V as well.

I/O error rarely happens but if it does it is likely due to hardware deficit.

Been through all the hardware. It's not that and by "All my PCI busses" I really mean all not just the slots. I've had this problem for a while. All and I really do mean all possible trouble shooting has been exhausted.

When I explained the system information readings I was getting and the programs I was using to collect that information as well as the trouble shooting steps I had taken and my conclusion. The Asus tech agreed and feels that the issue is almost certainly the motherboard. If it was just a hardware issue I would have fixed that long ago. I wish it wasn't the MB anything else would be less of a pain.

P.S. Your not wrong in stating:

I/O error rarely happens but if it does it is likely due to hardware deficit.

So if I sounded a little snotty I really didn't mean too, It's just that I've been through it all.