Splave’s How to Overclock Memory Frequency

Splave’s How to Bench Memory for Frequency, Featuring HyperX Fury Black

HyperX DDR4 Fury Memory

Extreme overclocker Allen “Splave” Golibersuch helps you reach new memory frequency heights using the HyperX Fury Black.

Benching memory for frequency is kind of interesting. Some love it, some think it’s useless. Manufacturers like it in general I believe because it is a simple case of a single number meaning everything. 4,000MHz is better than 3,000MHz. Our company can run at 4,000MHz, their company can only run at 3,800MHz. This is highly relatable to anyone, even someone with no knowledge of computers. They can relate to the fact that 4,000MHz is better than 3800MHz. They are less impressed by the tight timings that you worked so hard on, than they are of a simple 4,000MHz frequency. – Sorry, but it’s true.

HyperX HX426C15FBK4/16

Going on my own experience of benching Skylake platforms at Intel headquarters – l0udSilence and I were benching Hynix MFR super tight at 3,466MHz on LN2,  a feat that turned far fewer engineering heads than other guys running very loose 4000MHz+ validations. They simply like the high numbers!

So to conclude this introduction and finish my piece on why this “silly” benchmark is actually an important one to be good at, just look at all the HyperX competitions in history. They all have it. So why not give it a try?

Here is what I will be using in this ‘how to’ guide.

IMG_20160405_143245

  • Kingston HyperX Fury Black 2666MHz AFR
  • ASRock Z170M OC Formula
  • Intel Core i7 6700K Skylake Processor
  • Tsai Memory Pot and 8ecc Beast CPU pot.

IMG_20160326_232213

 

The key to getting nice high frequency results is patience, flexibility, voltage, and timings. If you don’t have the first two then you will struggle here. The world record by John Lam took a single session of more than six hours to achieve. Some CPUs like to be very cold, some like to be -80c and some like high SA won’t budge over a certain point.

VOLTAGES

Voltage here is a good starting point for you on LN2:

vCore 1.2v or lower

You will be running your CPU at x8 multiplier so you can run auto, or 1.2v is fine.

vCCSA 1.600v and higher

You will most likely find that unless your IMC is poor, that it will eat vCCSA until your CPU is dead.

I would suggest to start your vCCSA at 1.55v and push till you can’t raise the frequency any higher, then add little by little. I tested on an ES 6700k and 1.65v in software caused instant death. So there is a thin line between going to a higher frequency and the death of your CPU. I was at 1.625 vCCSA set in BIOS on another CPU and booting fine and the chip is still alive. So I would not go much higher than that, unless you see a reward for the risk!

vDIMM 1.8v

Pushing HyperX AFR, it is pretty standard to find a 1.8v DIMM is its happy setting on LN2 for frequency. You will find that adding more will not help and some amounts will even hurt your max.

vCCIO 1.25v

Unless your CPU has cold issues (with a stuttering mouse etc) you can leave this at 1.25v. If you have cold issues you can try as high as 1.4v but this will not help your memory frequency.

TLDR: Air max 1.5vCCSA 1.8vDimm // ln2 “safe” max 1.6vCCSA 1.8vDimm

 

TIMINGS

Here they are in all their glory. Again I’m not an engineer and this is all trial and error for me so you may find some that help you, but this is what I am using.

 

TEMPERATURES

CPUs will vary greatly. Some IMC’s don’t even scale on cold while others can gain full pot. Find what your CPU likes. I find around -100c to -120c to be nice and also convenient because most cold boot bugs are cooler than this.

Using a phase cooler can be convenient to help find what your memory likes temperature-wise, without wasting LN2 on the processor.

12325552_10201655595068424_1704380695_o

Kingston HyperX AFR loves cold and it’s beneficial to get it as cold as possible for what we are trying to achieve. Do whatever you must, use TIM on your memory pot, don’t use any fans right near the pot etc to help it get as close to -180 as possible.

IMG_20160405_142946

To break it down again we have these notes:

FACTS: TIMINGS, MEMORY VOLTAGE, MEMORY TEMP

Variables: CPU IMC Voltage Preference range (1.5-1.625v), CPU Temperature Preference (30c-180c)

We are ready to bench now so here is a normal scenario. We will go into the BIOS and set our timings and voltages and start to cool down our memory to as cold as possible and our CPU to around -100c while we are getting our BIOS setup. It is nice to push yourself here and pretend maybe you are at the HyperX OC Takeover finals in Las Vegas or other similar contests with specific time constraints.

  1. We are now at -50c on CPU and -50c on memory so we will try to boot at 4,133MHz at 100bclk.
  2. It boots, so now our game is booting with the bclk as high as possible at the 4133MHz divider.
  3. We Try to boot at 101bclk, it passes great!
  4. We try to boot at 103bclk, it also passes no problem.
  5. Now we try 105bclk and we get 55 hang on our bios POST code reader.
  6. Well we know our facts from above, we don’t need to change timings, we know maybe our memory needs to be cooler, and it could possibly be our IMC temp and vCCSA voltage. But, Since we have our facts lets try them first.
  7. We go back into BIOS and switch back to a 103bclk because it’s the last place we were able to train.
  8. We pull our memory down to -100c now and leave our CPU around -50c because we know AFR likes to be cold as possible.
  9. We try to boot 105bclk again and it passes.
  10. We try to boot 106bclk it fails again, so we go back to 105bclk and pull our memory down to -180c
  11. It passes training at 106bclk.
  12. We try 107bclk and it hangs on a 32 POST code
  13. Now we can start on our list of variables. ( Note: I always prefer to try colder temps before more volts).
  14. We boot back into a 106bclk and we pull our CPU down to -120c while our memory is still -180c.
  15. We try 107bclk again and it boot loops 3 times and shows us a failed OC message.
  16. Our CPU isn’t scaling with temperatures and with lower voltages so now we boot back in at 106 bclk and we are going to raise our vCCSA to 1.600v
  17. We set 107 bclk and it passes but we can see that we are getting less and less frequency and its getting more difficult to raise the bclk.
  18. Now we are going to switch to .500bclk raises at a time.
  19. We try 107.5 and it trains, we set 108.0 and it fails with a boot loop and failed OC.
  20. Now that we have tried everything and we can to train as high as we can, we will change from 107.5 in single .125 bclk increments to find our max trainable frequency.

At this point we haven’t even entered the OS one time. The key to all of this is training as high as possible. You are limited to about 2-4bclk raises tops once you are in the OS.

Now for the OS you are going to want a 32-bit Windows XP that is trimmed of most services but still has the ability to change bclk, obviously. Some tweaks that might help you are MaxMem between 150mb-250mb, depending on your services etc. Also, changing the XOC in CPU-Z seems to make file saving easier.

  1. Now we have found that 107.325 is our highest trainable bclk we will go into the OS.
  2. Open CPU-Z and in this case, the ASRock Formula Drive.
  3. Now we are going to raise the bclk in .250 increments until we find our max!

If you follow this well, hopefully you can have a strong result!

2207hyperx

Some pointers I have found. Try to limit mouse use and use hotkeys as much as possible. Sometimes you will have a lucky boot, and for reasons unknown, it will go higher. I have confirmed this with other top memory frequency overclockers. This also adds to the ‘patience game’.

IN CLOSING

Adding memory frequency benching to your overclocking skills arsenal will be a benefit to you. It will also aid in furthering the development of memory itself. Why not try diving into something new and challenging that may someday give you a leg up on the competition.

mhzhyperx

-Splave

324,993 total views, 526 views today

2 comments

  • Rod

    Great article. I struggle with the first attribute on your list and in turn have not spent much time on memory. However, this article definitely puts it into perspective and doesnt seem like to much of a daunting task. Thanks!

  • It consists of a lot of risk to regular users. I found myself in CPU won’t even start for a few days just to find it bootloop another day after memory overclocking from XMP 3333 1.35v to 4133 1.4v. Today it runs well but for how long? Maybe Skylake just like 1.2v on memory.

Leave a Reply

Your email address will not be published. Required fields are marked *