Haswell: 4.8GHz on Air.. Building A Haswell System

...and then add in this question:

DO YOU NEED THE IHS MOD AND WHY?

Any self-crowned technical wizard on the internet who posts:

"Do this mod, you wont regret it.. here is how its done but you do it at your own risk"

..and then does not preface that statement with what I am about to outline in this entire post is NOT anyone you should be taking technical advice from!

But before you personally commit to something like this there are few items you should consider first, and so you can make an INFORMED DECISION I will open this topic with the following information...

First, and foremost, although I have confirmed the first production run from one facility from industry sources that the Haswell 4770k does in fact include a very poorly designed IHS thermal connection between the top of the processor and the inside of the cap, that DOES NOT MEAN Intel did not split the design change and produce some with solder nor does it mean Intel wont switch up later and correct this on later steps of the same processor!

Just because I might have connections into Intel or in the industry does not mean that I get all the information. There are sometimes limits to information that my contacts have access to as they are not privileged in all facets of the business dealings nor are they involved with every aspect of production or decisions around production. I get good information but I also know that information comes with limitations which include non-disclosure issues.

Furthermore, although many of the Ivy Bridge designs are the same as Haswell, there have been confirmed differences in models, but again, there is no way to confirm all models were produced the same way therefore it is not possible to know if there is TIM or fluxless solder under all those caps. Intel appears to be experimenting and started that with IB.

My take on this is Intel pulled this stunt to cut costs since PC sales are down and they are watching the results. That means they could switch-up in production runs and completely return to fluxless solder, OR, decide that their flagship or hex core products have the better IHS and leave the standard models with TIM. There is no way to predict or know what they will do.

The problem with using temps alone to define if a issue may be present with a 6-8 core processor is a hex core will always run hotter than a quad, simple physics, so trying to guess if that chip has a TIM or solder can be a crap-shoot.

THEREFORE before you go committing to doing anything with a processor cap and removing it for inspection you should first consider the fact that there is a possibility you could open that cap and find out that solder was used instead of TIM, and second, you should ask yourself: Do I need this modification and if so, why.

To help you make that decision I am going to list facts here.. some people just do not clock or if they do they do not look for top-end, others like cooler temps regardless of the CPU speed or if they clock or not. Sill others want to clock but do not have the nerve to open a CPU cap be it a financial loss fear or lack of confidence in their ability to do it right. Because of all these differences you should know the facts before making the decision to crack that cap..

Unless you have far more 'cojones' than brains, NEVER crack a CPU cap (or make any other mods to hardware) before that CPU, motherboard, video card and memory have been fully tested for a period of time running unclocked.

We don't modify anything until we have fully installed a system all the way up through Windows, tuned, trimmed and verified with stability and defect tests.

To do so is outrageously ignorant as we have no idea when those parts arrive if the motherboard, CPU, video card or memory or any other components have defects.

Once the hardware is confirmed stable and free of defect, then we move to the next level.

Do I want to clock and if I do how high do I want to clock, and, is it worth it to me personally to run a high-end clock. If I do want to clock, what should I order for parts to clock Haswell?

Of course there are different levels of this. Lets look at each and what is required:

- Do not mod with the CPU or IHS   DONE!  If you prefer a better cooling solution then a lower grade liquid such as a Corsair H60 or a medium duty larger air heatsink will do the job.

 - Do not mod with the CPU or IHS. With the Asus motherboard you can after confirming the parts as defined in FACT 1, simply enter the ADVANCED BIOS settings under the AI Tweaker menu click the setting "OC TUNER and set it to "RATIO ONLY".

Do be aware that for best TEMP results a decent heatsink replacement would be best such as the Corsair H60 or a medium duty larger air heatsink will do the job but this automated clock can be done with the stock Intel heatsink you will simply run warmer on the Intel HSF but not dangerous. I would still suggest a medium duty replacement just to be sure and if you prefer cooler results. 

 The budget boards will most likely support this kind of low-end clocking.. you can check that at the Asus website or download the motherboard manual and review it first to confirm.

First, it is important to understand that at this point WE HAVE NO IDEA what quality of IHS thermal conductivity is in play, and, we have no clue as it if we have a processor that will run stable at LOW CPU voltage.

The SAFE BET is to always assume you will NOT GET a "golden chip" and you will receive a lousy one. 'Golden chip' processors are the processors that will run high CPU speeds on very low CPU voltage. They are few and far between! If you work from the assumption you will be in the worse situation for clocking when ordering the processor YOU WILL NEVER LOSE. It will only get better if you do happen to get to hitch a ride on the golden goose.

So under this scenario defined by the statement above;

- Do not mod with the CPU or IHS. In this case you will need a 'super-cooler' class air cooler such as the Thermalright Archon SB-E x2 or the Thermalright Silver Arrow SB-E Extreme  (or equivalent) OR you will need a liquid system that can handle the heat. With a air cooler you WILL most likely be limited to 4.3-4.5 and that assumes the chip will allow it!

BE VERY AWARE that some air coolers can cover the memory slots (The Thermalright Archon does not) and 'high-back' memory heatsinks wont fit under the fan! Some of those liquid systems on the market are no better at cooling than the large heatsinks I defined and they are far louder.

- This is a crap-shoot you will probably lose.

You will need to first clock your system up slowly and in increments to define if your IHS thermal connection is of poor quality or, assume that will be the case when purchasing parts but in this case you must provide yourself with the temp data in load testing. If you require 1.24 for 4.5 and 1.27 for 4.6 and your temps are very high using OCCT:CPU test alone, then you MUST take a lower CPU speed. 

Keep this in mind, the standard OCCT: CPU test WILL NOT LOAD that processor to the max. If you are seeing 80-85c max temps under that test then you ARE NOT loading that CPU with maximum instruction and you will hit 100c+ under the right conditions! You can be stable in OCCT but unstable in CPU load spikes.. that is a chance you will take without the IHS mod at those temperatures.

IF you can run OCCT: CPU test and not exceed 72-75c then your IHS is far better than many, and, that could be because you have a golden chip that does not require 1.27v for 4600. MOST processor will NOT run 4.6 at lower voltages and require 1.3 or above. If  you can pass that test at that temp then you do not need to mod the IHS and can run 45-4600 safely on air however I seriously doubt ANYONE will get to 4600 on air without the IHS mod even with a golden chip.

-Although you might get a golden chip and the Corsair H100 can deal with it, the ODDS ARE YOU WONT. Therefore if you are thinking of anything above 4.8 you need to plan to either BUILD a custom liquid system, (you better know what you are doing), OR, invest in a very powerful system that is far better rated that the latest Corsair H100 or H110! You will have to load test and I suggest you use the AIDA64 FPU Stress test alone to define your limits as you increase CPU speed and CPU voltage.

You can try it with a H100 or H110 liquid system but do not be disappointed if it does not happen. It might!   or it might not!

- I'm all in for a harem and palace in Saudi Arabia...  Keep dreaming...    NEXT!

-----------------------------

- In this case you will need a 'super-cooler' class air cooler such as the Thermalright Archon SB-E x2 or the Thermalright Silver Arrow SB-E Extreme (or equivalent). With a air cooler you WILL most likely be limited to 4.7-4.8 and that assumes the chip will allow it!

BE VERY AWARE that some air coolers can cover the memory slots (The Thermalright Archon does not) and 'high-back' memory heatsinks wont fit under the fan!

Choose your air cooler wisely and don't think for one minute a Cooler Master or a Zalman will cool that proc like a 'super-cooler' class heatsink will!

-Ok.. its your nickel,.. so then lets rock with a mod.

I have outline different scenarios above, but one thing is for certain, in all circumstances regardless of clocking, mods or not, you need to LOAD AND STABILITY TEST your system before making the decision. You may find out you just happened to get a golden chip, or, you may have received a processor that for what-ever-reason the IHS thermal connection is better than most.

BEFORE you crack a processor, make the INFORMED DECISION to open that chip! You may decide you like what you see on the screen at the clock you are able to achieve since Haswell is a FAR better chip that Sandy Bridge clock-per-clock. A Haswell at ANY clock speed, including default, is going to outrun a Sandy Bridge of equal clock!

@ 4.1 you can expect the same performance as a SB running 4500, no problem!

@ 4.5 you can expect the same performance as a SB running 5GHz, no problem!

@ 4.8 you can expect the same performance as a SB running 5.3+GHz, no problem!

HASWELL LOVES MEMORY SPEED AND LOW TIMING

REGARDLESS OF the CPU speed, you will want, at minimum, DDR3 2400 9-11-11-31 memory and if the sticks are decent quality they will run a 1T command rate with no additional voltages required (unless we are talking 5GHz+ then some voltage tuning may be required)

ABOVE 2800 - You are now getting into a grey area. YES, unlike platforms of the past, Haswell can run these speeds but stability will be based on the stability of the CPU and the temps and may very well require additional knowledge of DRAM settings in the BIOS for both sub-timing and voltage control. The only way to know is test it all in combat at the CPU speed you wish to run. The stick may run their default speed (in example DDR3 3000 12-14-14-38 for a fat wallet of 750 dollars  ) and not require any additional settings, OR, you may have to boost the DRAM Voltage, or DRAM POWER PHASE and DRAM CURRENT.

Once you exceed DDR3 2800 and are pushing clocks of 4800 and higher you are into a territory that can not be defined on a calculator prior to attempting such setups. The better boards should run those extended speeds with some adjustments.

THE BIOS SETUP AND NEW DANGERS WITH HASWELL

AT this point you may have it all under control as a experienced clocker and will figure the BIOS out yourself as well as a clock setup, but most people don't and order to know where you stand with a decision to mod the CPU you will need to know where the temps are in the clock you wish to run. To do that I will outline the basics to primary clock setups with this WARNING:

BE AWARE: Asus has changed how the CPU Voltage is used under the setting: "ADAPTIVE MODE" if you use that and you hit Haswell with full-on Haswell FPU instructions the BIOS is automatically programmed to add an additional voltage to what you defined and calculated as MAXIMUM.

This can be a disaster if you do not know exactly what you are doing so it is highly suggested that ANY clocking other than the simple automated 4.1GHz clocks established automatically by the BIOS that we ONLY USE: CPU VOLTAGE: MANUAL MODE and lock that voltage regardless.

My take: there are a lot of Titan users out there who got the 'shaft'  when the 780 was released. Glad I held off on that purchase.

I got rid of the silly LED lights in the fans as I have no use for or wish to look at that bling.

Please be patient as I am spread fairly thin right now with time. 

I am making it clear now that this is not a first year MIT Engineering class I am addressing and for those who are highly skilled in electronics or have a background in the subject should clearly understand that I will be addressing this topic on a level that most users can comprehend. I will be using terms and examples that are not 'technically accurate' to someone who may be educated in the subject in order to help those who do not have any idea what they are doing develop a far better grasp of the concepts presented which will in turn give them the confidence to establish a stable high performance clock with their system.

In other words, this is a Flight Sim hardware site and not a engineering lab. In that, I will be keeping this subject, the terms and examples VERY simplistic in order to help those who have no concept of electronics and how to get the most out of them.

Clocking Haswell: The Basic Rules and Introduction

Before I begin posting Haswell clock setups there are some basic rules everyone should understand. There are internet posted outlines about clocking Haswell with Asus motherboards and most of them are based on Asus data that was released by their reps. I have reviewed that information and found it to be accurate is some respects and not so accurate or at the very least 'iffy' in others.

CPU SPEED
MEMORY SPEED/TIMING
UNCORE or Northbridge SPEED

These are "The Three Kings" to clocking and each one of them is important. If one is out of whack, the rest suffer however Haswell has changed how one of them works.

This has always been the case however if you have ever read one of my discussions in the past about memory speed and timing then you know my take on this. The highest CPU speed is of course what we strive for but with FSX we do not want to take a 'major' hit in memory speed/timing (latency) nor UCLK Frequency (shown in CPUz on the memory tab as Northbridge Frequency). To do so means that 100Mhz higher tick up in CPU speed is being neutered by the system that feeds it data to process.

At the same time we would not want to take a 3-400Mhz hit on CPU speed to allow just a bit faster memory speed or timing. There is a balance between these systems that must be found and maintained.

Ok so now that you are completely confused let me simplify this for those who do not understand the terms I just used..

Think of the CPU as a customer who places a order for a product they intend to use as soon as it arrives. Think of the system memory as a warehouse in which a business stores its goods for processing/shipping. When a order comes in, if the warehouse is slow then it will take time to get that product off the storage rack and up to the loading dock.

Think of the Northbridge (which has changed over the last 5 years) as the "loading dock" where the goods are sorted for shipping out. If this is slow then regardless of how fast the warehouse gets the order to the loading dock, it will wait there to be shipped.

When you order something online you usually like to see it delivered quickly, right? Well, that order has to be processed, sorted and prepared for shipping, then it leaves the loading dock and travels to you. You cant use that product until it arrives. The CPU functions the same way, when it receives the data it can then move to process it.

All of these factors effect the final real world result with a system.. so we strive to make sure all of them are tuned and running efficiently.

MEMORY SPEED AND TIMING is the SECOND KING-

As I just pointed out above, how fast the memory works with the CPU defines how fast the data gets to the CPU and is recalled if needed. There are 2 aspects to memory, SPEED which is the bandwidth it runs, and, LATENCY, or memory TIMING which is the real key to making USE of higher bandwidth.

Think of BANDWIDTH as the size of the road and the number of lanes cars travel on.

Think of LATENCY as the STOP LIGHTS along the way and with respect to memory, every lane of traffic must stop at each light.

Now, what happens when you have a lot of cars even on road that supports many lanes and allows a large flow of traffic when they come to a stop light AND those lights are LONG?  Are they getting to their destination faster or slower?

SLOWER! OK, so the next time someone tells you that memory speed and latency DO NOT MATTER regardless of if it is a game or FSX, DONT argue with a fool, don't point them here and don't waste your time with these people...  simply tell them; as Bill Engvall says; HERES YOUR SIGN!

The design of Haswell has introduced memory speeds with stability that were never possible in the past. Eventually we will all be moving into DDR4 which what these developments are all about, memory speeds above 3000MHz. But regardless of that bandwidth, if the memory TIMING is high (or slow) then the number of WAIT STATES that occur along the way increase, effectively slowing that memory down.

This means if you use DDR3 2133 memory at a timing specification of 11-11-11 that you are effectively SLOWER than someone running DDR3 1600 at a timing specification of 6-8-6. Now, I know this question is crossing your mind; How do I figure out what is faster so I do not get ripped off with the parts purchase. There are formulas that will help define the BASE but without importing the speed and efficiency of the Uncore or Northbridge and the speed of the CPU into those equations you cant define it 100% and can get tripped up using the formulas. It requires a great deal more knowledge of engineering than we will be getting into here.

To simplify this I have defined BASE values for you to use when shopping for a Haswell system parts in the 2nd post of this thread. Its simple;

DDR3 2133 C9-11-10    Which you would not want with Haswell since it will run 2400
DDR3 2400 C9-11-11   This is what most users should go for as it is blazing fast
DDR3 2800 C11-14-14  This can be run but with even with Haswell remember that high CPU clocks can make any speed above 2400 difficult to obtain stable.
DDR3 3000 C12-14-14  NOTE: This is not easy to run at high clocks, stick with 2400-2800 unless you KNOW what you are doing with high-end clocks.

The C stands for CAS Latency.,. it is the primary number and the most important, but if those sub values are very high it effectively REDUCES the efficiency of the CAS value. This is where you must know what you are doing or you can be snookered by a memory company into paying hundreds of dollars for faster SPEED memory that is reality NOT FASTER. They do it all time..   example:

DDR3 1600 C6-8-8   this is SLOWER than DDR3 1600 C6-8-6 or C6-7-8

Last, always remember that the faster and lower the latency the more difficult it is to maintain HIGH STABILITY at faster CPU speeds. That is why we must find the balance where the memory speed and latency are running its peak performance with the CPU speed in operation. When clocking above 4.5Ghz always remember this:

In the past Uncore has been defined as 2 x MEMORY SPEED. This is important to the overall latency of the memory subsystem. i7 made this system far more efficient than it was before i7 hit the market.  Before i7 the northbridge was significantly more critical to performance and we used tricks in the BIOS to get the latency lowered at the northbridge so that 'loading dock' was shooting out orders at higher speed. Some of you may remember my lectures with CORE2 processors on the STRAP and use of engineering software to tickle that northbridge to lower the wait-states.

We no longer must run 2 x memory speed to obtain the latency result! We can now run 1.5x-1.6x-1.7x and 1.8x the memory speed and still be FASTER than i7 of the past. This is critical for TRUELY STABLE high-end CPU clocks of 4.5Ghz and above and the primary item Asus missed in their clocking setup outlines.

Of course the faster that speed the better, but let me give you an example here of just how efficient the Haswell system is with respect to this...

If you are running a CPU speed of 4800 and your memory speed is 2400 in the past we would want a UCLK or Northbridge speed of 4800 or 2x memory speed. With Haswell we can run a Northbridge speed of 4300 and be running FAR faster i7 than of the past.  Further, if the Northbridge frequency had to be dropped to 3900 from 4300, we only lose 1 nanosecond or LESS of latency in the memory system speed and we are still faster than i7 of the past!

That is an impressive GAIN in technology. It opens the door to very HIGH memory speeds with low latency and allows stability at high CPU clocks with little LOSS to the true speed of the system. I will say again, the faster this is, the better, but we no longer have to strive for or over-volt in order to obtain the result. Therefore the clocking outline with Haswell has been modified from guides of the past with respect to this setting in the Asus BIOS which is controlled by: MIN/MAX CPU CACHE RATIO.

THE DANGERS OF CLOCKING HASWELL

I have stated earlier that I can not define every Asus motherboard or especially motherboards that are not made by Asus. Therefore what I will outline during the clocking process is a way for YOU to verify your voltages and ensure they are within SAFE tolerances with respect to Haswell. I will outline the software to use and its version so you have the correct tools at your disposal for success.

With Haswell there have been some design changes with respect to current and voltage control. The design of Haswell itself can invoke automated increases in voltage past what you set in the BIOS regardless of those values being MANUAL or OFFSET.

Further: The use of ADAPTIVE voltage which is a feature in the Asus BIOS which Asus claims will allow far greater power efficiency comes with a very DANGEROUS caveat as it will allow a HUGE jump in voltage even if the adaptive offset is LOW.

In example; you could calculate and enter a adaptive voltage number that would place the total CPU VOLTAGE at 1.22v and if Haswell is hit with certain load criteria it will override that and you could be staring at a CPU VOLTAGE of 1.375 in the blink of an eye. Experienced users know that such a jump in CPU voltage could be deadly to a processor if the heat transfer system cant handle that voltage. You inexperienced users must understand that for each .10 volt increase in CPU voltage the heat goes UP and with Haswell it can rise significantly fast without the IHS modification that was discussed earlier in this thread and that will be outlined later to show everyone how to do that.

Therefore it is my recommendation that unless you are exceptionally trained in electronics and know how to read and define the increase that will occur under certain LOAD conditions that all users stick to CPU CORE VOLTAGE: MANUAL ONLY and not use offsets or adaptive to control Haswell at higher CPU clock. Be aware that even with manual voltage control there will still be automated increases in voltage, just no where near as much.

With that being said I think its time to start outlining a BIOS setup for clocking however you MUST REMEMBER that because of the confirmed issues that exist in the thermal transfer contact point between the CPU die and the cap that covers it THAT WITHOUT A MODIFICATION TEMPS WITH HASWELL WILL BE VERY HIGH AND CPU CLOCKS THAT REQUIRE MORE THAN 1.20 VOLTS ARE GOING TO NEED SUFICIENT COOLING TO ACCOMPLISH THEM SAFELY.

This section assumes you have read every post in this thread and have followed the guidelines established for setting up a proper Haswell clock.

It is important to note that regardless of CPU and memory load tests or how long they are run, overclocking can still produce errors. There is no such thing as a 'totally stable' high-end overclock when 3D render is added to the load. The ODDS are far better after extensive load tests confirm there is no exceptionally unstable issue present and if a user knows what they are doing they will probably not see any issues however be very aware that regardless of personal experience you can pass every CPU/memory load test with flying colors and STILL not be 100% stable under 3D render graphic load conditions which may require speed or voltage adjustments to trim out.

If a user is not well versed in overclocking this can also be true of LOWER end clocks. This is true of any system overclock and not just Haswell.

I am going to address ALL clocks setups from 4.0GHz through 4.8GHz with Haswell. I have developed a routine for this that will be nearly universal for all clocks in this range. In doing so it has simplified the process by quite a bit.

I will not be addressing clock speeds of over 4.8GHz. There is a reason for that. In order to obtain such a clock you will have to be running a far better cooling solution than the typical Corsair liquid system can provide, and, the processor you received would have to be one of the top in its class for low voltage/high CPU speed. Since both of these factors will be RARE I will be keeping the clock setup outline to a 4.8GHz max speed.

If you are intend to push the CPU higher, and you actually KNOW what you are doing, you can use the same information I post here for max temp/max voltage and without exceeding those defined limits attempt a 4.9-5.0GHz clock, but be aware as of the date of this posting no one is running a 24/7 safe voltage clock and stable on Haswell @ 5.0 and above on air or liquid. Those speeds are not a realistic goal with Haswell, 4.8 and below is, assuming the processor will allow it.

INFORMATION YOU SHOULD KNOW:

The first time we clock Haswell and before considering modification to the thermal contact point between the cap of the processor and the processor die (or IHS mod), the first order of business will be to establish where your individual Haswell processor falls for voltage and clock speed and at what temperature. We want to establish that the clock we wish to obtain can be achieved and if not how high the system will go and for those who are considering a IHS modification, to define if you really need or want to modify that processor thermal connection point which I will outline in another part of this thread.  

Asus has outlined their own approach to clocking Haswell. They define the memory should not be XMP set which runs it at it lowest speed and highest timing, and instead clock the processor as high as possible first and then set the memory to XMP. I disagree with this approach as it sets the user up for a CPU speed their memory and UCLK can not follow without excessive voltage and current. This is unrealistic to what a user would want to achieve and is in my opinion designed for marketing purposes over real world use. Instead, my approach establishes the memory manufactures speed/timing for the base overclock of the CPU which is what a user will run in the real world.

We will be enabling C-States in the system BIOS, changing them and using that system to lower the CPU temp at idle. This goes a long way in extending the life of a processor when it is not under full load and is why the Haswell certified PSU is needed.

Haswell has two primary voltage/load states in which the top-end voltage/load temperatures will probably not be seen unless a software product that is 'AVX' instruction enabled is run but that does not mean we do anything FOOLISH such as use ADAPTIVE CPU Voltage. In this day and age and with a Flight sim tower you will most likely never hit Haswell with AVX instructions except with certain load test software which we will use to define the top-end temperature and stability of the clock, but that is not to say at some point a software you use may not introduce those instructions, therefore we NEVER use Adaptive CPU voltage when clocking Haswell because regardless of the 'calculated adaptive value' under the right conditions a CPU voltage 1.35 and GREATER will be invoked, instantaneously and without a warning.

TWO VERY IMPORTANT FACTORS TO CLOCKING HASWELL

HOW FAR CAN YOU GO

Lets assume you have the best liquid system on the market money can buy....

-First and foremost the QUALITY of the processor in the manufacture process that you received. You may find your Haswell will clock 4.8Ghz on a manual locked 1.35v and fully load test stable, or, you may find your Haswell processor may require 1.45v to run 4.8GHz with all load tests and run stable. The same is true for lower end clocks, you may be able to run 4.5GHz @ 1.18v or it may require 1.32v or more!

Unlike Sandy bridge there is an exceptionally large window of voltage vs. clock speeds with Haswell. Mostly due to the unreliable HIS thermal contact in the manufacture process and also driven by the far higher clock-per-clock performance increase in the design of the processor over processors of the past where Haswell gets hot, FAST.

This means it is impossible to define an 'exact' CPU VOLTAGE starting point for all systems, NOR is possible to estimate where YOUR Haswell will clock and at what voltage, however the method I post below should get you to the point where you know the limit and where your system stands without a lot of hoops and ladders to jump through.

-Second, the quality of the IMC in the processor will define how well UCLK will run in relation to the speed. The memory speed and the number of sticks installed can also influence that because the higher we go in memory speed and the more memory sticks in the system, the less stable that area of the processor can be. As CPU speed rises more voltage is required and that voltage requirement goes UP with the number of sticks and the faster the memory speed.

I posted that DDR3 2800 is the highest stable recommended speed for memory, however that does not mean you will be able to run high UCLK if your Haswell CPU IMC happens to be a bit finicky. That is why I defined 2x4GB DDR3 2400 9-11-11-31 is a better bet for users who intend to try for speeds greater than 4.5GHz and are not tech very savvy with electronics.  

QUICK REVIEW BEFORE STARTING:

Synopsis:

Going back to what I posted earlier in this thread, "The Three Kings", with Haswell that is how we will approach clocking the system. The first order of business is to stabilize the memory controller with the memory installed and run the manufacture speed and timing of the memory, and then establish a CPU speed/voltage that is stable below a defined temperature limit. From there we increase the UCLK frequency (or Northbridge frequency as CPUz displays) and establish its upper end, once those two are locked and stable then the process of reducing voltages starts with repeat load tests till the stable point is found at the lowest voltage settings.

Memory tweaking IF it can be done is always LAST after all software is installed. Memory clocking is a PERK, not a requirement and if you purchased the right memory most users should simply let the memory do its job at the manufacture rated speed with perhaps one minor tweak.

- Decide what CPU speed you intend to try and run for the initial tests: I addressed this in the 2nd post of the thread and explained very clearly what you need in the way of a CPU cooling solution to try and obtain clock goals. I will RECAP here again so you do not attempt anything foolish starting out and without the IHS cap modification.

*4.0-4.2GHz - The stock Intel heatsink MAY allow this but the system will most definitely run hotter on that heatsink. If you replaced the Intel heatsink with a Corsair H60 or a larger air heatsink, these speeds will work for you at far lower temperature.

*4.3-4.5GHz - ON AIR you will need a 'super-cooler' class air cooler and ON LIQUID you will need a Corsair H100/H110 or better quality liquid cooler. On the air super-cooler your limit may very well be 4.4GHz and REMEMBER even with liquid cooling if your Haswell is not a good clocking chip you may not even get past 4500! This All COMES DOWN TO: The luck of the draw.

*4.6-4.8GHz - ON HIGH END LIQUID ONLY WITHOUT THE IHS MOD and be very aware you may not be able to achieve either speed if the CPU requires more voltage than the average Haswell processor.

In all cases Intel Hyper Threading will be DISABLED. Hyper Threading serves NO PURPOSE with FSX and if you desire the system run with Intel Hyper Threading enabled you will most likely have to accept a lower clock speed/CPU voltage or exceptionally higher temperatures.

This clock requires high-end liquid cooling only without a modification to the CPU IHS contact point. Even high-end liquid can FAIL this clock without the CPU IHS mod. Be aware that exceptionally dangerous temperatures are possible with this clock outline as well as the potential to push maximum voltage specifications.

If you have read through or worked with the outlines as presented thus far you may have noticed that with Haswell as we increase CPU speed as well as cache speed with it, the larger the gap between one level of clocking to the next in voltage range requirement enters the equation.

In this next outline and assuming the CPU is not a 'dud' for clocking, you are going to be faced with what are known as the 'walls' in clocking where one user may find their CPU must go from 1.28v to 1.42v in order to obtain a single step up in CPU speed with the cache speed following reasonably behind.

As with all other outlines we start with the cache speed set to a fixed lower speed and then proceed to increase it once our CPU speed is tested stable and within temp specs.

Rest assured, this clock is not easy to obtain and the CPU mod I will outline later opens the door to this level of clocking far better than working with the current Intel design Haswell chips that all appear to have poor connections between the CPU die and the inside of the cap that covers it.

Do not be disappointed if your system can not achieve this clock. A 4.5GHz clock with a CPU cache speed of 4300-4400 or 4500 is an exceptionally excellent clock with Haswell and remember at a Haswell CPU speed of 4500 the system is easily running in a comparable 5GHz range with SB processors.

HOW TO FIREWALL HASWELL

-or-

BRING UP THE HOLY HAND GRENADE! <--clicky

I would suggest starting at 4600, then progress to 4700 and then finally, try for 4800. Any one of these clocks will be a superior clock setup.

CPU CORE RATIO: PER CORE        make SURE this is set:  PER CORE

Set each of the next 4 boxes to: 46 for 4.6GHz, 47 for 4.7GHz or 48 for 4.8GHz

Highlight the setting with the mouse, enter the value in the box and hit ENTER on the keyboard after each numerical entry. BIOS settings will not accept unless you enter the value and hit ENTER

You should have each core (4 lines) set now to: 46    (or 47 - 48 if you prefer)

We are going to run a bit higher cache with this clock than others to start off...

MIN CPU CACHE: 41

MAX CPU CACHE: 41 

Note: Sometimes this requires the MAX value be changed first then the MIN

We keep the CACHE speed lower for now and after we establish that the CPU speed has tested clean then move to increase the CPU cache speed.

NOTE: it is nearly impossible to run a CPU cache speed above 4500 with a CPU speed of 4600 and greater without exceeding SAFE voltage LIMITS. Consider yourself LUCKY if you can run a cache speed of 4400 with a CPU speed of 46-4800.

You have now established a constant 4600MHz (or 4700-4800MHz) CPU, 4100MHz NB Frequency

Scroll down and set: CPU CORE VOLTAGE: MANUAL MODE

Select the setting below and input the following value:

FOR 4.6GHz: CPU CORE VOLTAGE OVERRIDE:   1.34    <--- type this number and hit ENTER

FOR 4.7GHz: CPU CORE VOLTAGE OVERRIDE:   1.37    <--- type this number and hit ENTER

FOR 4.8GHz: CPU CORE VOLTAGE OVERRIDE:   1.39    <--- type this number and hit ENTER

NOTE: My starting voltage value range for a 4.6-4.7-4.8 clock is:  1.30 - 1.43. I am splitting the difference here to start out. 1.34 - 1.37 - 1.39

Be VERY careful with these next values and NOTE THE DECIMAL POINTS!

CPU CACHE VOLTAGE: MANUAL MODE

CPU CACHE VOLTAGE OVERRIDE:   1.30

CPU SYSTEM AGENT OFFSET MODE SIGN: +

CPU SYSTEM AGENT VOLTAGE OFFSET:   .300

CPU ANALOG I/O VOLTAGE OFFSET SIGN: +

CPU ANALOG I/O VOLTAGE OFFSET:   .050

CPU DIGITAL I/O VOLTAGE OFFSET SIGN: +

CPU DIGITAL I/O VOLTAGE OFFSET:   .100     

CPU INPUT VOLTAGE:   1.88

---------------------------------------------------

NOTE: ROG BOARDS THAT DEFINE 2 CPU INPUT VOLTAGES:

INITIAL CPU INPUT VOLTAGE:   1.88

EVENTUAL CPU INPUT VOLTAGE:   1.88

---------------------------------------------------

Hit F-10 SAVE AND EXIT.  REBOOT back into the BIOS, enter the AI Tweaker menu, scroll down and CONFIRM every setting listed above starting with CPU CORE RATIO: PER CORE  and ending with: CPU INPUT VOLTAGE:  1.88

Check you input values directly. If there is any setting that is not right or has changed, reset it as defined above in the outline for your clock speed then hit F-10 save and exit, reboot and verify in the AI Tweaker list in the BIOS again. Once all settings are confirmed, then begin the testing process by booting into Windows. One of two things will happen:

a. Your system will boot into Windows just fine and remain stable while you load up the test software. If this is the case please proceed to the next section: Haswell and Overclocking 104 - Stability and Load Testing Software and follow the directions.  (if you have already reviewed section 104 you may move directly to section 105)

b. Your system will lockup or crash booting into Windows. If this happens, (hard reset if required) boot into the BIOS and raise the CPU core voltage .01volts.

Example: If you set CPU CORE VOLTAGE OVERRIDE: 1.34, then raise that to 1.35, F-10 save and exit and try booting into Windows again. Continue raising the voltage by .01v until you have a successful boot into Windows and appear stable. Once you have obtained that goal please proceed to the next section: Haswell and Overclocking 104 - Stability and Load Testing Software and follow the directions.  (if you have already reviewed section 104 you may move directly to section 105)