Guide for custom stepped i7 2600K "Per Core" overclocking = extra performance!

HardLuck

n00b
Joined
Feb 19, 2011
Messages
14
Guide to setting "Per Core" Turbo overclock range of 4.5 to 4.8Ghz using individual "Per Core" Turbo Ratios

Disclaimer
If you attempt the overclock described below, it is entirely at your own risk. The author assumes no responsibility for any damage caused to your system, should you attempt it.

Author's Note
The following guide is written for an intended audience of new to moderate overclockers. I don't espouse to being an expert overclocker with the Sandy Bridge architecture but wanted to share my overclocking experiences and achievements, with the view that others may also be interested in this particular overclock.

Introduction
If you're a moderately cautious i7 2600K overclocker like myself and use a good air-cooling CPU heat-sink, you may have set your overclock to a maximum of 4.5Ghz across all cores, and are happy with that speed. Much of the 2600K overclocking community regard this Turbo of 4.5Ghz across all cores to be a safe 24x7 overclock for daily computer usage. At 4.5Ghz, CPU core temps generally stay below 80 degrees Celsius at full load, and is an easy overclock to achieve in the BIOS (most settings need only be set to default/Auto).

However, with some BIOS setting changes you can gain more performance out of your 2600K for apps that don't utilise all four cores, and without risking higher temps. Many applications do not utilise all four cores at one time, and that leaves extra headroom in your CPU.

What I've done is instead of setting my overclock to be 4.5Ghz across "All Cores" I've set a "Per Core" stepped overclock. Now when an app only uses one of the four CPU cores it automatically Turbos up to 4.8Ghz. When only 2 cores are used: 4.7Ghz, for 3 cores: 4.6Ghz, and for apps that use all four cores: the original base 4.5Ghz. In my testing, the temps and volts of the higher overclocks (when less than four cores are utilised) is less or equal to that when all four cores at 4.5Ghz are stressed at full load. That means bonus performance for us 4.5Ghz overclockers in day-to-day computer usage. Of course the CPU still idles at 1600Mhz, and only jumps to the specific stepped Turbo clock when the CPU is under load.

My computer
CPU: Intel i7 2600K
Cooler: Hyper 212+ (push & pull fans)
Motherboard: Asus P8P67 Pro (original B2 stepping)
BIOS version: 1305
RAM: 4x2GB Corsair DDR3-1600Mhz C9
Primary Hard Drive: Samsung SpinPoint F3 500GB HD502HJ
OS: Windows 7 64-bit
Graphics: nVidia GTX 460 1GB

The BIOS settings which worked for me for this Per Core overclock
Note: all BIOS settings are left at their default values unless specified below. I've added some comments about my findings during my testing:
AI Overclock Tuner: X.M.P (or Manual)
BCLK/PEG Frequency: 100.0
Turbo Ratio: By Per Core
1-Core Ratio Limit: 48
2-Core Ratio Limit: 47
3-Core Ratio Limit: 46
4-Core Ratio Limit: 45
Internal PLL Overvoltage: Disabled
Memory Frequency: (as per your RAM specification)
EPU Power Saving Mode: Enabled
EPU Setting: Auto
Load-Line Calibration: High (or Auto, however Auto will cause higher voltages and temps. I found any less than a setting of High for this "Per Core" overclock causes a BSOD during stress testing)
VRM Frequency: Auto (or 350)
VRM Spread Spectrum: Disabled
Phase Control: Standard (or Optimised)
CPU Voltage: Manual Mode (I found CPU volts and temps went too high during stress testing if this setting is set to Auto for this overclock. However a setting of Auto is fine if all cores are set to 4.5Ghz or below)
CPU Manual Voltage: 1.360 (At this voltage the actual voltage and watts used will still fluctuate depending on CPU load, as can be seen using Real Temp 1.37. While the CPU voltage of "1.360" works best for me , your CPU may require 1.365 or maybe more/less. If you get a Blue Screen of Death with error code 124, that typically means your CPU voltage is too low. Always watch your temps when trying a new CPU voltage. My own limit is to ensure the hottest core stays under 80 degrees Celcius, using air cooling with room temperate in the mid-20s Celcius. Each to their own liking, though.)
DRAM Voltage: (as per your RAM specification)
CPU Spread Spectrum: Disabled
CPU C1E: Enabled
CPU C3 Report: Enabled
CPU C6 Report: Enabled

Some Screenshots of my BIOS:

1305_1.gif


1305_2.gif


1305_3.gif



Screenshots of only one core being stressed:

Real_Temp.gif


Task_Manager.gif



Screenshots of all cores being stressed:

Real_Temp2.gif


Task_Manager2.gif



Tools I recommend to test and monitor your computer with these new settings:

Real Temp v3.67 (Shows core temps and real-time clock speeds. Tip: click the button on the top-right hand corner to cycle between time, volts, and watts)
Prime95
Windows Task Manager (Use the "Performance" tab to see each of the 4-cores/8-threads working.)
Cinebench 11.5 (Can test all cores or single core performance. Choose File > Advanced Benchmark)
AIDA64 Extreme Edition v1.60.1321 beta (I use Tools > "System Stability Test" > "Stress FPU" only, which causes the most heat, volts and CPU load of any stress test I've ever used. This is a commercial tool, but a trial version is available. Play with the new CPU Mask option to test only selected cores/threads. You'll need to convert a binary number of 8 character length (representing 8 threads) into hex. For example: To test the first and last threads of your 2600K CPU, the binary number 10000001 equates to 81 hex. See screenshot below )

CPU_Mask.gif


That's all folks. Hope you've enjoyed my guide, and are able to gain extra performance out of your new i7 2600K CPU! :)

- HardLuck
 
Last edited:
Nice guide. It seems very easy to follow.

I wonder how this applies to a 2500k?
 
Nice guide, but if you're only getting 4.5ghz out of a 2600k across all four cores you're doing something wrong. I'm at 5ghz easily with mine, and can bench at 5.2. Just in a Antec Kuhler 620 as well with the stock fan.
 
Thanks for the guide HardLuck,

But whenever I try these single core tests, the work is spread out among all the threads. I assume this is due to Hyperthreading.

So I assume you have Hyperthreading turned off then?

If so, wouldn't the advantages or hyperthreading be greater than the extra .3 GHz single-core performance? (unless all your apps are only single core aware)

If not, how do you get only one core to work in Cinebench 11.5 for example? When I select Advanced and Single Core, it spreads the work out.

Regarding CPU Voltage: I have found that using CPU Voltage Offset mode instead of Manual helps keep the idle temp down.

The optimum Offset value needs to be found through some trial and error and varies with the CPU.

I have found that my optimum Offset is +.010 for 4.6GHz OC (all cores).

With that, at idle, the CPU voltage drops to 1.008v. Idle temp drops to 27 vs. 31 at the higher manual voltage setting.

Under load, voltage using the AIDA64 stress test it is similar to yours varying between 1.344 - 1.360.
 
Last edited:
Thanks for the guide HardLuck,

But whenever I try these single core tests, the work is spread out among all the threads. I assume this is due to Hyperthreading.

So I assume you have Hyperthreading turned off then?

If so, wouldn't the advantages or hyperthreading be greater than the extra .3 GHz single-core performance? (unless all your apps are only single core aware)

If not, how do you get only one core to work in Cinebench 11.5 for example? When I select Advanced and Single Core, it spreads the work out.

Regarding CPU Voltage: I have found that using CPU Voltage Offset mode instead of Manual helps keep the idle temp down.

The optimum Offset value needs to be found through some trial and error and varies with the CPU.

I have found that my optimum Offset is +.010 for 4.6GHz OC (all cores).

With that, at idle, the CPU voltage drops to 1.008v. Idle temp drops to 27 vs. 31 at the higher manual voltage setting.

Under load, voltage using the AIDA64 stress test it is similar to yours varying between 1.344 - 1.360.

Hi MarkyM,

Thanks for your reply. To address your questions, Hyperthreading remains turned on. You are correct that some benchmark apps spread the "single core" load over multiple cores, such as Cinebench 11.5 (and also Prime95), when only testing single threads. What this does in practice is that you get the same turbo boost to 4.8ghz for single core usage, but the temps and load is balanced over a number of cores, which Windows regards as more efficient overall from a real world computer usage perspective. However, to "force" stress testing individual single cores/threads, I used the CPU Mask option in AIDA64, which I mentioned in my guide. That gives total control over which combination of cores to stress test, allowing you to discover the various loads and temp limits per core, to ensure you have achieved full stability and remain in a good temp range. I looked and couldn't find an alternate free tool which could stress just a single thread like AIDA64 could. (If you find one let us know). But if you have Real Temp open during real-world usage of your computer, you'll notice the Mhz bouncing up and down the whole time, going into stepped overclock when it needs to.

Finally, I appreciate that some prefer to find and use an optimum Offset voltage instead of a fixed one. My gut feel is that the Offset voltage may not scale as well on a stepped per core overclock due to the various volts that each step requires. However I stand to be corrected on that, as I've not tried it myself. On my system with a 1.36 Manual voltage, the Real Temp VID shows the volts drop to 0.97 at 1600Mhz (idle) with around 5 watts of power. Therefore my temps remain cool at idle, with low power usage, so I'm happy to remain with the Manual voltage setting for now.

- HardLuck
 
OK, thanks.

Understood now that Aida64 is the only app that can fully stress one core at a time.

I have not tried your stepped core scheme yet so that makes sense as well.

--Mark--

Hi MarkyM,

Thanks for your reply. To address your questions, Hyperthreading remains turned on. You are correct that some benchmark apps spread the "single core" load over multiple cores, such as Cinebench 11.5 (and also Prime95), when only testing single threads. What this does in practice is that you get the same turbo boost to 4.8ghz for single core usage, but the temps and load is balanced over a number of cores, which Windows regards as more efficient overall from a real world computer usage perspective. However, to "force" stress testing individual single cores/threads, I used the CPU Mask option in AIDA64, which I mentioned in my guide. That gives total control over which combination of cores to stress test, allowing you to discover the various loads and temp limits per core, to ensure you have achieved full stability and remain in a good temp range. I looked and couldn't find an alternate free tool which could stress just a single thread like AIDA64 could. (If you find one let us know). But if you have Real Temp open during real-world usage of your computer, you'll notice the Mhz bouncing up and down the whole time, going into stepped overclock when it needs to.

Finally, I appreciate that some prefer to find and use an optimum Offset voltage instead of a fixed one. My gut feel is that the Offset voltage may not scale as well on a stepped per core overclock due to the various volts that each step requires. However I stand to be corrected on that, as I've not tried it myself. On my system with a 1.36 Manual voltage, the Real Temp VID shows the volts drop to 0.97 at 1600Mhz (idle) with around 5 watts of power. Therefore my temps remain cool at idle, with low power usage, so I'm happy to remain with the Manual voltage setting for now.

- HardLuck
 
Per-core settings don't work at all on my P8P67-M PRO B3, BIOS 703. Just seems to be stuck at 3.7GHz no matter how many Prime95 threads I run.
 
What this per core overclock is basically doing is replicating the default Intel Turbo Mode behaviour as per the CPU running at stock speeds. You'd be aware that at stock settings, a 2600K runs at 3.4Ghz but can Turbo up to 3.8Ghz for mostly single core apps. What I'm basically doing is replicating the stock Turbo feature, but starting with the higher base overclock of 4.5Ghz instead of 3.4Ghz. 4.5 is the base speed I'm most comfortable with, while others prefer to go higher if their system can support it. The stepped Per Core overclock mostly benefits general computer usage for apps that don't use all 4 cores, then reverts to the base overclock when benchmarking/using all four cores at once.

t2uciet: Are you checking your VID voltage at idle with Real Temp 3.67? Different tools report different things. Best to ensure your watts and temps remain low during idle, which Real Temp can also report in real-time. At full load my CPU can use over 100 watts, whereas at idle it still drops to around 5 watts whether I'm using a Manual or Auto voltage.

Blahman: I'd recommend to ensure that no background threads are running during testing. Perhaps even temporarily unplug/disable your LAN connection to ensure no network activity during testing. Other than that I'm not sure - it could be due to differences in BIOS versions between different mobos. We'll wait and hear from others with the same mobo as yourself.
 
Blahman: I'd recommend to ensure that no background threads are running during testing. Perhaps even temporarily unplug/disable your LAN connection to ensure no network activity during testing. Other than that I'm not sure - it could be due to differences in BIOS versions between different mobos. We'll wait and hear from others with the same mobo as yourself.

Yes I'm sure nothing else was running, and I always stability test with minimal processes and services and no network connection, and anyways there is no other reason it would be running at 3.7 GHz than if it was reverting to stock, which is my suspicion. The multis I had set up were 46, 44, 43 and 42. I have seen other funkiness with this BIOS and B3 boards in general so I would not be surprised to hear it is a BIOS issue.
 
Update: Advising users not to yet update to the new BIOS version 1502. Further stress testing with an Asus P8P67 Pro motherboard has caused new BSODs under 1502 (using the settings in this guide), but which worked fine and stable under 1305. It may seem 1502 wants higher voltagages along with higher LLC and Phase settings compared with 1305, just to be overclock stable.
 
Last edited:
Hello. I know this is an old post, but I can't find any answer that is close to what I have as far as hardware. I wanted to OC my setup but am having issue. The pc boots up to the Windows logo and hangs. It my settings because if I revert back, windows loads find. Here is my hardware. What would be cause it only loads to windows logo and freezes?

Asus P8Z77-pro
intel i7 Core 2600K
ram: sector 5 (4x4gb)
ssd: main drive
 
duceduc,

Try setting your 2600K CPU to no more than 4.5Ghz across "All Cores" and leaving most of the BIOS settings set to Auto. If Windows still won't boot then all you can do is to reset all BIOS settings back to their defaults, change one setting at a time, and test booting into Windows after each change. The Per Core stepped overclock in this thread is quite tricky to get stable, so save your time and just go Auto, would be my advice to you.
 
To add to what I've mentioned above, I think there is something wrong with my current bios firmware. I may have been a bad flash or incomplete. Looking at the ss below, the me version is missing and my ram memory freq is not displaying.

I would hate to have to re-flash the mb. The last time I did it, I somehow messed up and I had to get another mb (this one).

3s7VHw.jpg
 
Glad you revived this thread. I never tried this approach to my 2500K. I used the all cores max approach. (still kicking 4.4 )

Kid
 
@hardluck

I finally found out there was an issue with the firmware 2104 I was in. I have reset cmos and reinstall the same firmware. I am able to boot into Windows one time and when I reboot, I cannot. It get stuck in Windows logo with the hard drive light led and somethings it doesn't.

I am thinking it's a bad firmware version. What version are you running to get this setup working?

I am currently stable at 4.5 running on all cores. I want to try setting up the per core method.
 
Nice guide, but if you're only getting 4.5ghz out of a 2600k across all four cores you're doing something wrong. I'm at 5ghz easily with mine, and can bench at 5.2. Just in a Antec Kuhler 620 as well with the stock fan.

Ok. I am back trying this once again. Followed the steps in the first post. The only thing I didn't change was the CPU C3 & CPU C6. These have to be disabled or I am unable to boot through the Windows logo. The issue I am having is that all cores are registering at 4.5MHz even though I have set it to per core at 48,47,46,45. Is there a step I am missing? Is this setup limited to certain MB? I am using the Asus P8Z77-V PRO with the latest firmware v2104.
 
Back
Top