LGA 1156 Core i7 & i5 Overclocking Guide

[andy5174]

This guide was written based on Core i5-750 CPU and Gigabyte GA-P55A-UD5 M/B!

This is for peoples who want to OC their PC but lazy to comprehend the basic knowledge of overclocking!
Thus, please don't criticize it for its unprofessional way of writing.

Warning: Overclocking DOES void the warranty. Neither TOM Hardware nor I will be responsible for any damage caused by overclocking!


0) HSF & Thermal grease

1) Decision making on the OC

2) How to start?

3) Voltages tweaking

4) Torture test


0) HSF & Thermal grease

- CoolerMaster Hyper 212 Plus has great performance for the cost.

- Prolimatech Megashadow is one of the top heatsinks which is paired with 2x 2000rpm 19dB CoolerMaster SickelFlow Fan here.

- Arctic Silver 5 is an excellent thermal grease with reasonable cost.

( Click here for the guide on thermal compound application! )


1) Decision making on the OC

What is the best OC?

Performance-efficiency wise

- The maximum OC that can be achieved near stock core voltage(1.20V shown on CPU-Z at 100% load).

- Power consumption increase is in LINEAR region! (The power increase in an exponential form after 3.6GHz -> Bad efficiency!)

For LGA1156 CPUs, 3.6GHz OC satisfies these two conditions.

Performance ONLY

The maximum frequency that can be achieved within the absolute maximum core voltage(1.55V for LGA1156 CPUs).

To be safe, you want to be within the maximum core voltage which is 1.40V for LGA1156 CPUs.

LGA1156 CPU Documentation from Intel!


2) How to start?

1. Enter the BIOS by pressing Delete key during POST(Power On Self Test)/Boot Screen.

2. Set BCLK, CPU multiplier/ratio, QPI Clock multiplier/ratio and Memory multiplier/ratio so that you can achieve the frequencies you want.

BCLK x CPU ratio = CPU frequency
BCLK x QPI ratio = QPI link speed -> set the ratio to the lowest possible value!
BCLK x Memory ratio = Memory frequency

My settings are:
CPU frequency: 180x20=3.6GHz
QPI link speed: 180x32=5.76GHz
Memory frequency:180x8=1440MHz

3. Manually set the memory timing according to the specification of your RAM, leaving everything unspecified as Auto.
(You have to set the DRAM Timing Selectable to Quick or Expert in order to do this in P55A-UD5!)

4. Disable the Turbo Boost Technology!

5. If it's impossible to get your system stable, disable C1E, C3/C6/C7 and EIST.

(C1E, C3/C6/C7 and EIST are power saving features and CPU will always run at highest clock when ALL of them are disabled.)

For i7-8xx CPUs, disable HT (Hyper Threading Technology) as well under this situation.


3) Voltages tweaking

Immensely useful information from Intel Datasheet! Give it a read before you move on, because it is going to help substantially in the following steps!


PCH: Intel P55 chipset voltage

- main I/O interface
- display connectivity
- audio feature
- power management
- manageability
- security
- Storage feature

It is pretty obvious that the higher the CPU frequency, the more frequent the I/O(input and output) and so the higher the power consumption of "main I/O interface". Hence, the PCH voltage needs to be increased based on the same current(P=VI). However, 1.10V should be sufficient for any OC achieved on air cooling.

PLL: Phase Locked Loop voltage

- Processor/IMC(Integrated Memory Controller)/other internal clocks
- Clock multiplying of processor is provided by an internal Phase Locked Loop

Generally, 1.8V(default value) and 1.9V for OC under 3.6GHz and 4.2GHz respectively.

Vtt

- L3 shared cache
- memory controller
- processor I/O power rail

Since memory controller frequency is dependent on BCLK, the higher the BCLK, the higher the Vtt required.

1. To achieve the absolute maximum OC, begin with applying the Maximum voltages(to be safe) or Absolute Maximum voltages(at your own risk) in BIOS.

(LGA1156 CPU Maximum/absolute Maximum voltage by Intel: Vcore= 1.40/[strike]1.55[/strike]1.40, Vtt= 1.155/[strike]1.21[/strike]1.40, Vram= 1.575/[strike]1.65[/strike]1.80, PLL= 1.89/1.98) with PCH=1.10

For i5-750 ONLY: The following settings are good points to start with!

In order to find out the minimum stable voltages, keep lowering the following voltages ONE at a time until your system becomes unstable.

(For example, varying ONLY the Vcore with all other voltages FIXED while trying to find out the minimum stable Vcore.)

The lower the voltages are, the cooler and healthier the CPU is!


3.6GHz:(24hrs Small FFTs test, 30hrs-2mins Large FFTs test and 14hrs-12mins Memtest86+ stable!)
Load-Line Calibration: Enabled (or Level2 for Gigabyte GA-P55A-UD5)
Vcore= 1.20V (CPU-Z idle)
QPI/Vtt/IMC= 1.14V
PCH= 1.10V
PLL= 1.80V
RAM= Specified voltage for your RAM

3.8GHz:
Load-Line Calibration: Disabled
Vcore= 1.264V (CPU-Z idle)
QPI/Vtt/IMC= 1.149V
PCH= 1.10V
PLL= 1.85V
RAM= Specified voltage for your RAM

4.0GHz:
Load-Line Calibration: Disabled
Vcore= 1.38V (CPU-Z idle)
QPI/Vtt/IMC= 1.25V <- Higher than absolute maximum Vtt
PCH= 1.10V
PLL= 1.88V
RAM= Specified voltage for your RAM

4.2GHz:
Load-Line Calibration: Disabled
Vcore= 1.43V (CPU-Z idle) <- Higher than maximum Vcore
QPI/Vtt/IMC= 1.30V <- Higher than absolute maximum Vtt
PCH= 1.10V
PLL= 1.90V
RAM= Specified voltage for your RAM

2. Set BCLK=210 with CPU Multiplier=20, (210,20), OR BCLK=200 with CPU Multiplier=21, (200,21), to make the CPU 4.2GHz which is the maximum on average chips.

3. Check for stability of your system:

- Stable: Jump to the next step, 4) Torture test, for thorough stability test. OR Set the BCLK and CPU Multiplier to (215,20 ) or (205,21) to get higher frequency.

- Unstable: Up the voltages at your own risk. OR Lower the BCLK and CPU Multiplier to (200,20) or (190,21) to be on the safe side. Keep repeating this until your PC is stable.


4) Torture test

1. Software required: CPU-Z, HWMonitor, Core Temp, Memtest86+ and Prime95

2. Run at least 12 hours Memtest86+ test and Large FFTs test of Prime95.

Monitor 100% load core temperatures

Make sure that your core temperatures are ALL within 72.7C which is the thermal specification from Intel!

p.s. Under 18C ambient temperature, my 100% load core temperatures are 48C-44C-44C-43C at 3.6GHz with the cooling solution specified earlier.

Prime95 Torture Test Options

- Small FFTs: CPU Stability test
- In-Place Large FFTs: Overall system(CPU/RAM Interface) stability test
- Blend: Memory stability test, NOT very useful! Use Memtest86+ instead

Click here for Prime95 instruction!

Go back to 3) Voltages tweaking and up the voltage a notch if it's not stable!

*Tip*: The best way to find out which voltage is causing the instability is setting the voltage that might be too low according to your 6th sense to its maximum value with ALL other voltages staying unchanged and see if the system is stable after this. If it is stable, then it's the voltage you need to tweak. If it is not, try tweaking another voltage in the same way.


For i5-750 ONLY: You might be extremely unlucky to have a chip that is much worse than the average with the stable voltages being much higher than those values given by me. What I would do under this situation is setting all the voltages to their maximum(NOT absolute maximum) values and lowering them ONE at a time until you get instability. However, try it again with C1E, C3/C6/C7 and EIST disabled before doing so.
(p.s. Maximum/absolute Maximum voltage by Intel: Vcore= 1.40/[strike]1.55[/strike]1.40, Vtt= 1.155/[strike]1.21[/strike]1.40, Vram= 1.575/[strike]1.65[/strike]1.80, PLL= 1.89/1.98)


N.B. Any OC below 3.8GHz should be very easy and don't go above that if either of your core temperatures or core voltage is already near its maximum specification at the point.


Good luck!

UPDATES:
28June10:
- Absolute Maximum Vcore, Vtt, and Vram change in latest Intel datasheet revision!

 

[EXT64]

Do you have a link to the newest version. Looks very nice, if a bit complex:

http://www.overclock.net/intel-cpus/645392-how-run-linpack-stress-test-linx.html

The efforts we got through to get our CPUs to fail :lol:

 

[EXT64]

Nevermind, I guess this is the most recent:

http://www.xtremesystems.org/forums/showthread.php?t=201670

 

[andy5174]

Yes, 0.64 is the latest.

 

[wolfram23]

I've got my CPU stable at 3.3ghz (160x21) and it's good enough, however I spent a few hours trying to get it to stabilize at 180x20 and at times I had it running ok on the large FFT (Prime95) but I could never get the Blend test to work. Temps were just barely into 80 on one core. I think it must have to do mostly with the RAM, but I'm making sure it's all the manufacturer's recommended settings (8-8-8-24-2T, 1.65V) and in fact underclocked to 1440 since the next option was 1800. There's so many other timing settings tho it's confusing.

About temps tho, can someone explain why I have one core running up to 10 or 11C cooler than the rest? For instance in a burn test, core 0 @ 75, 1 @ 80, 2 @ 68, 3 @77. Is it like a super core? Or broken? Or maybe the heat sink has best contact at that spot...? Seems wierd to me. (Using both HWMonitor and Core Temp)

 

[EXT64]

Yep, heatsink. Either minute differences in contact, pressure, etc. Also, I imagine the core near the uncore/mc might get a little warmer, though not sure on that.

Blend is a primarily memory test.

 

[EXT64]

Ok, just finished the LinX Test.

To confirm everything was working (as per the guide I posted a few posts up):

Max Theoretical GFlops: 57.6
Proportional Estimate off of i5 at 4GHz: 53.1-54
My Best Run: 52.4
My Worst Run: 51.9

So I'd call that a success, very close to estimate and not even that far off of max.

Temps: Between 59 and 55 Celsius, so clearly a hair higher than Prime95.

Completed 50 passes, problem size 20014, 3072 MB RAM, in 1 hour 39 min 10 sec with no errors. Time for each run was around 102.2 seconds.

 

[RJR]

Now, see if your computer can handle something a little different. It's just a super pi on steroids called y-cruncher:

http://www.numberworld.org/y-cruncher/

Oh yeah, I wouldn't go over 100M (about 30 seconds) until you see what kind of temps you are getting.

 

[EXT64]

I'll give it a try (eventually).

Have you seen significantly higher temps with that (vs. prime or a linpack)?

 

[RJR]

Have you seen significantly higher temps with that (vs. prime or a linpack)?

It does get hotter than LinX, hence the 30 second trial run time. You can check out some runs at XtremeSystems, Poke has been keeping track of some.

 

[andy5174]

Ok, just finished the LinX Test.

To confirm everything was working (as per the guide I posted a few posts up):

Max Theoretical GFlops: 57.6
Proportional Estimate off of i5 at 4GHz: 53.1-54
My Best Run: 52.4
My Worst Run: 51.9

So I'd call that a success, very close to estimate and not even that far off of max.

Temps: Between 59 and 55 Celsius, so clearly a hair higher than Prime95.

Completed 50 passes, problem size 20014, 3072 MB RAM, in 1 hour 39 min 10 sec with no errors. Time for each run was around 102.2 seconds.

Impressive! Very impressive! I will try the latest BIOS and see if I can get similar results as you.

BTW, the core temps could be even higher if you choose to use all the free RAM by pressing "ALL" next to the Memory(MiB) selection column.

 

[EXT64]

It was pretty close to all already. I ran it for a few seconds beforehand just to clear the memory out. I think though there were still a couple hundred MB's doing nothing.

I never OC'ed before F8, so no idea if it helps.

My i5 is a fairly old one (I think perhaps of the original batch).

 

[EXT64]

Hmm, not sure what I was doing wrong, but y-cruncher seemed to barely get up to Prime/LinX test temperatures. Should I be using the benchmark or stress tester?

 

[andy5174]

IMO, stress test.

 

[EXT64]

I had tried both last night with similar results. I'll have to read up sometime to see if I am doing something wrong.

 

[EXT64]

Slight Update: Core Voltage is down to 1.15V. Passed 50 LinX passes with about the same (52 GFLOPS) performance. Tried 1.12, however that failed on the 15th LinX pass.

I haven't checked to see if there is a measurable drop in power consumption. Load temps are down, however that is due to much lower ambient today.

 

[andy5174]

A lucky draw there you have!

 

[EXT64]

Yep, big step up from my FX-60 that could barely do 2.8 GHz (+200 MHz) on stock volts :lol:

I'll have to run some more power tests soon.

 

[wolfram23]

LinX is cool. Just ran a quick test, but I musn't have the settings quite right or too many background apps running because I only got around 35.6-35.7 GFlops at 3.6ghz.
Cool program tho I'll have to try some more that was just a first "hey let's see!" trial lol.
EDIT: Ok did a real quick retest after exiting all programs and gadgets, and got 51.4GFlops. Much better. Looking forward to playing around with it some more.

 

[EXT64]

Yep, that sounds about right. I like that program. It works well if you treat it right.

 

[wolfram23]

UPDATES:
28June10:
- Absolute Maximum Vcore, Vtt, and Vram change in latest Intel datasheet revision!

I just noticed this today in looking at the spec sheets. I'd love to have some discussion on this as it is a pretty significant change. I made a post (http://www.tomshardware.com/forum/284958-28-voltage-specs#t2131680) about it... now somehow the i5 750/i7 8xx can support MORE Vtt than an i7 9xx?? I'd love if someone can clarify or whatever.

 

[Mark Heath]

Great guide, thanks.

Shouldn't this be a discussion thread (not a question thread) though?

 

[EXT64]

Since it is an easy mistake to make, a lot of threads are posted accidentally under the wrong category.

 

[Mark Heath]

But mods can change the categories between discussion and question.

 

[cappo619]

Sorry for the 2 year bump, but I am going to try this tonight. I hope this works on a P7P55D ASUS Mobo

 

[chrysalis]

I think I have an aweful i5.

Here is my story.

I built the rig, and my previous rig was a core2duo which was probably the most stable rig I have ever owned, in addition the core2duo overclocked with ease on stock clocks.

After rig built I proceeded to install windows and use it, this was on stock voltages, clocks etc. Ram was overspecced, faster rated than speed I needed.

I started getting random BSOD's 90% of them were during multimedia video clips etc.

Long story short I had to overvolt the ram and imc to stop the BSODs on 'stock' speeds. Seems the voltage I used on the imc is the max safe voltage stated by intel, so no headroom for overclocking. Anyway I ran like this for ages and eventually changed the ram.

The new ram had lower rated voltages so the voltage was lowered but I left the imc as it was and still had the ram overvolted as I didnt want the bsod's back. system stayed stable.

Today decided to overclock albeit temporarily, followed this guide for the 'safe' 3.6ghz overclock. 1.2 vcore etc. Left my imc overvolted at 1.15. Believe it or not this easy 3.6ghz overclock I had a BSOD whilst watching youtube, to me you cant beat multimedia video clips for stress testing. On the original bsod's I had it had passed several days worth of prime95 etc. but playing videos was a good way to trigger BSOD's and seems still is when hardware over pushed. So now I am back on the default relying on turbo mode for on demand light overclocks to 3,2ghz. either I got very unlucky with my chip and its a dud, or this gen of cpus suck for overclocking. Made even worse than if I waited for sandy bridge those cpus seem a bliss to overclock like the core2duos were.