Thinking I would go for the 10900K and then, once available, an RTX 3080 or 3080Ti.
Not really stressed about finishing the build right now so I can wait for the GPU, take my time and build it, test, overclock, test, watch the grass grow, and so on in the meantime. Plus, I will need a new monitor too, and pick a PSU (Looking at Corsair RM850x for now) and drives since the new PC will be in another room than the Sandy-Bridge E rig I have now, which I will keep for other stuff. So lots to think about. I think I won't push higher than an all core clock of 5 GHz, so not really going to break any records. I will probably be getting the Corsair iCUE H150i RGB PRO XT cooler for the CPU. 

But, I'm kinda stuck on choosing the memory. So I've read that it's more important now than in the passed to look at the memory QVL of the board so I have been looking there and on the GSKILL RAM Configurator for the board to have some more options since the motherboard QVL is kinda limited.

https://www.asus.com/Motherboards/ROG-MAXIMUS-XII-HERO-WI-FI/HelpDesk_QVL_Memory/" rel="nofollow - https://www.asus.com/Motherboards/ROG-MAXIMUS-XII-HERO-WI-FI/HelpDesk_QVL_Memory/

https://www.gskill.com/configurator?page=1&cls=1529635169&manufacturer=1524725352&chipset=1589508048&model=1589509007" rel="nofollow - https://www.gskill.com/configurator?page=1&cls=1529635169&manufacturer=1524725352&chipset=1589508048&model=1589509007

So I'm looking at 32 GB in total and there are 2x16GB and 4x8GB kits. There seems to be more choices when looking at 4x8GB kits though, so maybe that is the way to go? What do you think? 

So here are some of the kits I have found on either the motherboard memory QVL or over at GSKILL or both:

F4-4000C15Q-32GTZR (4x8GB, 1.5v)

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232973" rel="nofollow - https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232973

F4-4000C17Q-32GTZR (4x8GB, 1.35V)
https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232675" rel="nofollow - https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232675

F4-4000C19D-32GTZR (2x16GB, 1.35V)

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232768" rel="nofollow - https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232768

F4-3800C14Q-32GTZN (4x8GB, 1.5V)

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232934" rel="nofollow - https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232934

F4-3600C14Q-32GTRSB (4x8GB, 1.45V)

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232894" rel="nofollow - https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232894

F4-3600C16D-32GTZR (2x16GB, 1.35V)

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232905" rel="nofollow - https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232905

F4-3200C14Q-32GFX (4x8GB, 1.35V)

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232623" rel="nofollow - https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232623

F4-3200C14D-32GTZ (2x16GB, 1.35V)

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232206" rel="nofollow - https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232206

I don't know if 4000 CL15 or 3800 CL14 would be pushing my luck...and they run at 1.5V. 

The other kits run between 1.45V and 1.35V. Would 1.5V be too much in the long run?
The 3600 CL14 is rather interesting as well. I will have to say that the 2x16GB kits don't really impress me much, when compared to the potential performance of 4x8GB kits here. The 2x16GB 3200 CL14 is on par with the 4x8GB kit though.

Appreciate your comments.

If that's the case, I always go the best performance option which is 

or this

If you are not willing to go there then its either...

F4-4000C17Q-32GTZR (4x8GB, 1.35V)
https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232675" rel="nofollow - https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232675

Only the ignorant who turn it on because of other issues and think it has made a difference, and then spin the affinity mask wheel of fortune.. believe it is. Then there is the expert 'stuttering Steve' at AV** who cant get his story strait.

That is a significant difference and one the user can easily see the value in running a lower clock due to temperature with HT enabled. 

Someone would have to show me visually on the screen without silly graphs and statistics how HT is significantly increasing the performance of any current iteration of FSX or P3D before I would give up physical core clock speed for virtual threads. In fact in many cases it creates more stutter issues that must be tuned out using a setting that does not exist by default in the config files.

Your assumption is wrong. The issue from the manufacture is getting a 16GB (instead of 8) quality high speed/low latency stick to run stable with a window.

It's all a give and take and this is where you run into the voltage deal. You want the memory amount and the speed but you don't want to run 4 slots.., you have to deal with it. 

So if you can put 2 16GB high speed/low latency sticks in a 4 slot board the board will clock the CPU with less headaches over 4. You have to deal with the cost and the voltage if that is what you want with a full-on manual clock.

There is no getting around it in specs.

Here is how I look at it..  unless you are really good at overclocking, you are not going to get above 5GHz stable and secure anyway, so take the route that nets the best memory performance support, eat the voltage and enjoy the system.

If that isn't something that appeals, then go for the lower voltage options but be very aware in a full-on clock, you may very well not be running 1.35 with those either.

Mine are rated for 1.35 and I must run them @ 1.45 24/7 on a 8 cylinder 24/7 5.2GHz overclock. I also manually tuned the secondary/tertiary timings too. No problems here. Tower is properly ventilated and the ambient room environment is 100% controlled @ 78F (or lower) x 35-50% humidity max - 24/7 year round.

The environment a tower is designed to run and lives in under conditions is something many do not think about. The room could be great but if the tower is shoved under a desk, or, if the outside wall the tower is close to gets hit by the sun and varies conditions... and,, the user has not thought about the heat trapped..  it changes the entire dynamic for component longevity. 

The more memory per slot and the lower the timing with higher speed, the higher the voltage

The more slots populated and the lower the timing with higher speed, the higher the voltage

The more slots populated, the lower the memory timing and the higher the CPU clock, the more CPU and supporting voltage required.

We are are getting into 32GB, something we never had to consider in the past for a typical system. With that comes all the caveats in droves no matter how many sticks.

Therefore and in my opinion, keep the clock reasonable. Go for the communication performance and don't worry about the number of slots or the voltage. If you are concerned about that then give up the quest for the high end memory communication perf and go with what you are comfortable with.

32GB and over throws everything under the bus (no pun intended) for a performance clocked system. 

Unless the memory manufacture comes across a bin run of chips that will run low voltage at max perf (and believe me, they will get top dollar for the product) then you have to give and take based on what you are comfortable doing.

Those sticks are probably going to get fairly warm no matter what you choose if its above 3200 and low latency so make sure the tower is properly designed for ventilation, is not shoved in a closed corner or under a desk where it can't breathe or circulate fresh air and the room it is in doesn't see temps in the 80F+ range... it should be fine.     should be  

 

What surprises me though, is that not a single 2x16GB kit found during my search runs at 1.5V. I don't know if that makes them too hot or what, but that is my observation. Anyways, I will probably throw in the towel here for the search for 2x16GB kits and get either the 4x8GB 4000CL15 or the 4x8GB 3800CL14 and see where it gets me. And I will not focus blindly at a CPU clock of 5 GHz. I mean, if I can run the memory as specified, I will accept a lower CPU clock if needed.

But you also noticed the voltage difference between the two and therein lies the answer along with memory/motherboard manufactures who know folks who are going to run to the 2 stick option are going to either want a later 64GB option, or, a high CPU clock. In either of those cases they don't want to push the envelope to the edge on 2 modules.

I would need to get my hands on a set, but I would like to bet with the right chip selection those 16GB modules could run higher speed/lower latency on a higher voltage, just like the 4 stick counterparts, but then that same voltage is being applied to 2 enclosed modules.. 

These manufactures also know most folks are not the sharpest tacks in the box and to avoid returns and complaints they would opt to go the safe route till they rethink their approach. Remember, this 32GB+ thing in clockers/gaming has not been a norm, is just coming around to main stream today, the memory manufactures have a stock to get rid of on top of redesign of product down the road for that purpose.  

However, like I said I would like to bet they could be run past their spec but that would be someone who wants to play and doesn't expect to get a refund if they won't. I would look at the chips on the sticks, something most would never ever think of doing, before making the decision to try it. Knowing the bin-run of chips makes a huge difference with success with something like that.

Biker eh?   I had to retire mine some years back. You should have gone to Sturgis, SD! Best music and cleanest jails in the Midwest! 

Hans,

I was in the same predicament as you. My current case is old Antec 193 with a 200 mm side fan. It has been solid, reliable, and cool ever since I built this system in 2013. 

https://www.newegg.com/black-antec-performance-one-series-atx-full-tower/p/N82E16811129176" rel="nofollow - https://www.newegg.com/black-antec-performance-one-series-atx-full-tower/p/N82E16811129176

The reason it shows out of stock is because it isn't made anymore and I couldn't find any other cases with a big side fan that I liked. It seems that people are more interested in light shows than good cooling these days. I would have reused the Antec P193 but it doesn't have enough room on top for a 360 mm radiator.

I instead settled on this Fractal Design 7 XL case which is a new model, has plenty of room for radiators, and can be purchased with metal side panels to allow installation of a large side fan if desired. I have an EVGA hybrid cooled 1080ti and there is room to mount its radiator on the back side.

https://www.fractal-design.com/products/cases/define/define-7-xl/black/" rel="nofollow - https://www.fractal-design.com/products/cases/define/define-7-xl/black/

I can easily install the 200 mm fan from my Antec P193 into the 7 XL if I need to. The extra bonus for you is that Fractal Design is based in Sweden.

Thought you might be interested,

Ted 

Unfortunately it seems memory for the most part comes with those silly lights today as does the Corsair AIO liquid systems. I don't mind the Corsair software to control the PSU (its needed to switch it to single rail anyway) and the AIO waterblock head lighting as it works well but I have found the Asus motherboard software as well as the memory manufacture's software is terrible for efficient design and hits the CPU far too hard in some cases. I just uninstalled all that junk and let the memory sticks run their default light show. 

Heh, I remember a game a long time ago where you rode a bike and I think the end goal was to reach Sturgis. And that's probably as close to Sturgis I will get  
Oh here it is:  https://www.youtube.com/watch?v=sVh9MlfzkFY" rel="nofollow - https://www.youtube.com/watch?v=sVh9MlfzkFY

I still have my Suzuki RF900R which I bought new in 1998, and my friend had one also and that was why we both were on that mailing list back in the day.

Interesting thoughts about playing with 2x16GB kits. It did cross my mind trying to clock a 1.35V 2x16GB kit but I'm not sure if I ever would reach the performance of that 4x8GB 4000 CL15 kit I'm thinking about, or if the sticks are even nearly binned as high as those in the 4000 CL15 kit. And as you say, the heat would be distributed on two sticks instead of four so I don't know. 

But I think a 2x16GB kit as this one would be a candidate for a clocking attempt:

F4-3200C14D-32GTZ (2x16GB, 1.35V)

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232206" rel="nofollow - https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232206  

Or maybe this:

Looks like there are upper vents on the side door that allow intake..  those would feed the radiator without drawing warm air up through.

Be aware... 

I would call that a project box as it is not a high end built tower to exacting specifics like Ted pointed you to. If you don't want deal with a project box, go with what he outlined which is nice .. there may be some metal tweaks and tangles that need to be addressed including slot fits but its a heck of a lot cheaper than a few hundred bucks. I like large towers even if the blinky light gamers hate them...  they breathe with low RPM intake fan airflow!! 

Except the rear fan..  that one I would bump up just a little more in speed under full load (like 1500-1700 RPM full) compared to the intakes which will run 600-900 max and to compensate for the volume coming in which will balance out with the 3 AIO H150i fans 

I run 230mm on the side with the perf metal in place (didn't cut it) and 140mm on the front and rear. (cut the back perf metal out) no air filters. Don't need them in my environment, just clean the tower/fans and video card once a year. Blow out the radiator and the PSU with compressed can air. PSU runs a bottom fan.  

Suzuki RF900R    Rice Rocket!   nice  

I was a saddlebag schlep/cruiser.  

32GB+ is a give and take all the way round. 

Being where you are don't see a problem with heat on the sticks if the tower and room location is right.

As for the Roxwell case, I can't find it here so I'm concidering either the 750D or the R7 and make the modifications as planned. And for storage 1x Samsung 970 1TB M.2 NVMe and 1x Samsung 970 2TB M.2 NVMe. As for the PSU, would you say there is advantage to get one that I can switch to single rail? The one I looked at can't be switched (Corsair RM850x) so maybe I should look at the HX850i?

The RF was a budget sports/sports tourer from Suzuki, manufactured for the US market between 1994-1997. For Europe, it was available between 1994-1999, the 99 model being extremely rare. 

Yes, I would do the HX850i (requires the USB link to the motherboard) and use the corsair link software to lock it in single rail mode.

The bike? Its still a rice rocket!..  much respect for being able to run one of things! 

I always tried to have the queen seat filled in order to lean back comfortably while cruising with mine.   

Hell, sometimes it even came with a massage. 

Heh I don't recommend taking passengers on the RF. Having someone on the back seat on my bike means that person would likely be leaning on me...But having the rear case attached to the wingrack gives them a natural backrest and can rectify the issue 

Seems like you had some really nice pillion riders with you 

I miss those days..  I know I can't handle a pig anymore but I've thought about renting a tricycle and hitting Sturgis one more time. Problem is I don't need a damn silver alert sent out but it would fun just mess with some heads too. 

As for the tricycle, there is a saying that goes something like this: You don't regret the things you did, you regret the things you didn't do.

 I've also read that I don't really have to use the Corsair software for the CPU Cooler iCue H150i unless I want to change the silly RGB lighting, which I don't care about anyway.  The fans and pump can be connected to the motherboard instead and I can use the Asus software to control them instead of some potentially CPU eating corsair software. 

The water pump on the H150i does not change speed on their product and always remains a constant 1080-1100 although it can be monitored. Probably best to do from the AIO water pump input on the motherboard and set a alarm for RPM failure. I use the Asus software for tower fan control. As for lights on the waterblock, I set those to a basic green-yellow-red based on temp 30c-60c-75c in the iCUE software. I have found the software has an update to it almost monthly but there is never any setting lost during or after a update.

This is the first time I have used liquid for anything other than experimenting and the first AIO. I am not displeased with it in any way however if it ever goes out I probably wont waste time getting it fixed or spend to replace it and will instead go back to some type of air solution.  

As for the PSU..  that is up to you. As long as its a good 850 and has single rail ability that's all that matters. 

Just got off the phone with a Corsair rep..   this info is not in any of the provided nomenclature with the product I have had for over a year and it does change some things, not considerably but still in a positive direction.

You can set custom fan curve speed as I mentioned and save it I set mine to 26c for the low end and 38c for the high which ramps the 3 fans after placing them on that profile. The temp selections work for the 78F room environment.. yours may be different.

To change the water pump speed, one of the settings on the side must be selected first: Quiet, Balanced, Extreme THEN click the water pump RPM readout and that will change the pump speed to a fixed value. 

Quiet (default) = 1100

Balanced = 2100

Extreme = 2800

There is a slight noise change going to extreme but I noticed little to none switching to balanced. 

It is a static change for speed. The fans will still be controlled based on liquid temp but the pump remains static at the speed set. 

I am going to run some compare tests here and see where the CPU hits its highest temp @ WATT load to find out where or if there is a difference from past test results.

I'll know more in a day or so.

I was able to achieve 5.3 stable and within temps on my specified load tests, not a problem

Problem was the voltage to do it. 

So, I can run 5.3 and based on my calculations I could be replacing the slug in 25-40 months, or.. go for the fallback plan.

5.2GHz just like I had except I raised the CPU core voltage slightly (I use adaptive and a very special BIOS setup) and was able to get 200MHz more out of the cache (originally 4600, now 4800) and win for the long haul.

So I am now @ 5.2GHz and 4800 with a cooler overall CPU.

Liquid systems are all about the liquid temp..  every single C you can drop them, the better.

So mounting the radiator and fans attached to it outside the tower, on a side where no heat can be reintroduced.. may not be a bad idea but where you are may not make a difference.

None the less..  setting the pump to BALANCED = 2100RPM..   is a win

I would not run EXTREME = 2800RPM   like any other mechanical device, you don't put your foot flat down on the accelerator 100% of the time and expect the device to run the long haul distance.

So before installing Windows I will run the memtest86 v8.4 with XMP set and then each stick 3 full test passes. That's clear.

Or do we do this once a clock is established? Or do we use that other software you mention in the Flight Simulator 11 thread (HCI Design Memtest) once Windows has been installed?

And thank you for presenting the Overclocking after Haswell outline over in that thread.

ha!.. hope you are enjoying the outrageous shenanigans going on in that thread 

I wrap up the entire outline in one post.. probably delete the original first part and repost it all in one. I'll link the new location where parts 1-3 are now.

as for memory testing..

One stick at a time. New processors error correct with more than one, so if you test more than one you could get fooled by the system.

Its done to test your memory purchase first.

After the offline is done, you really should obtain the paid version of HCi Windows Memtest as it has become a very inexpensive a gold standard for checking memory in Windows and especially in a clock, but you must also understand that when running that test it must pass the 500% point in order to call it a solid setup and then later, a stable clock with memory speed/timing.

Some things have changed..  so once you have confirmed with the offline (single sticks) the first time, if you are using HCi Memtest, you do not need to repeat the offline again unless you wish to do just for peace of mind. Before HCi that really wasn't an option.

And yes, the Flight Simulator 11 thread is interesting in many ways  

And the outline couldn’t have come at a better time 

Fly happy wrote: Got it. The bible thread however had an update mentioning DDR4 with the download link to memtest86 v8.4 which still seems to be the latest version so all good. Okay so I test each stick separately offline and then put all four sticks in, install Windows and then test with the paid version of HCI Memtest in Windows as described. And yes, the Flight Simulator 11 thread is interesting in many ways   And the outline couldn’t have come at a better time

I'm losin it! 

Don't know if it will help you but here is a link to the PDF for the guide in the forum 

http://gex.flight1.net/forumimages/Overclocking_9700K.pdf" rel="nofollow - http://gex.flight1.net/forumimages/Overclocking_9700K.pdf

https://www.techradar.com/news/nvidias-rtx-3000-gpus-need-psu-adaptor-and-a-staggering-amount-of-power-going-by-leaked-photos" rel="nofollow - https://www.techradar.com/news/nvidias-rtx-3000-gpus-need-psu-adaptor-and-a-staggering-amount-of-power-going-by-leaked-photos

Of course that is their top dog card, but still..  the others can't be too far behind.

Might be better to spend the extra 20-25 bucks for the HX1000watt if you intend to throw down the gauntlet and grab their top end 2080ti replacement, which is suppose to have like 20-24GB of video memory too. (at least that's what I have read)

This just got dumped out there and there is still nothing in stone.. the 850 is probably just fine being a 80+ platinum but thought I would let you know.

I do not know if they are going to ship those cards with the proper adapter which I understand is 2 PCIe ports, except wired differently?  So keep that in mind too.

I sure hope there's some adapter being supplied with the cards if they need one.

Found this video:

So I should probably install Windows 10 to a GPT partition, after following these settings recommended by Intel? 

https://www.intel.com/content/www/us/en/support/articles/000025535/memory-and-storage.html" rel="nofollow - https://www.intel.com/content/www/us/en/support/articles/000025535/memory-and-storage.html

I did some reading about the GPT partition which seems to be associated with UEFI,and someone claimed you gotta have atleast 2 partitions in which one has to be EFI or System Reserved partition of minimum 100MB and another one is Primary Partition for Windows Folder...

This is getting complicated and I'm probably just overthinking stuff.

And here they say FOUR partitions will be created (Step 14) https://www.tenforums.com/tutorials/1950-clean-install-windows-10-a.html" rel="nofollow - https://www.tenforums.com/tutorials/1950-clean-install-windows-10-a.html

I would just make sure it is the only drive in the system when Windows is installed so nothing is installed away from the drive itself on another drive.

There is probably a NVMe support setting and well as the EUFI settings in the BIOS too under drives and/or under BOOT

Here you go...try and beat the price.

https://www.bestbuy.com/site/ibuypower-gaming-desktop-intel-i9-10900k-16gb-memory-nvidia-geforce-rtx-2070-super-8gb-1tb-ssd/6419490.p?skuId=6419490" rel="nofollow - https://www.bestbuy.com/site/ibuypower-gaming-desktop-intel-i9-10900k-16gb-memory-nvidia-geforce-rtx-2070-super-8gb-1tb-ssd/6419490.p?skuId=6419490

I have a question about if I set up BIOS outlined by Intel in my last post, do Windows automatically set the drive to GPT during installation or do I need to manually do it in diskpart before I start the installation? Remember, I have the 1 TB Samsung 970 Evo Plus NVMe drive. I have now planned to have the entire space of the drive unallocated and let Windows make the partitions it needs.

OS Type [Windows UEFI mode]  perhaps reads like that in your BIOS

One thing I would suggest and this may not be as much of a factor to you because of your geographical location and the typical ambient temps, but I have always opted for a video card that expels heat out of the tower through a shotgun (get it out), over these 2-3 fan open face video card coolers that tend to disperse heat everywhere.

To me, those are fine for a typical tower but for a tower that is intended for clocking performance, get the heat out and make sure it can not recirculate. You don't want any more heat rising and hitting the HSF or a radiator.. pump it out the back 

I'll look for a video card with that solution once they are available. 

I can sort of see it with a standard heatsink. It will dissipate some heat through the solid backplate mount itself but with a liquid cooler head, I don't see that happening. 

But do have fun and do let me know the result of the experiment 

Haven't finished the build yet but it's close.

When I wrote that thread over 7 years ago I had run into that situation so many over the years I made a standard procedure to install the chipset driver from the DVD/CD first, then update it to the latest. It eliminated any guesswork or possible issues.

I have noticed the quality of the installers have been better over recent years however you are right, to be sure of ignition without any gotcha's, run the disk chipset driver first, then after reboot update it. 

Good catch, nice to see someone reads and retains! 

When it comes to this subject today, run the chipset installer from the DVD/CD and also check to see if there is any other driver listed on the disk that is not listed at the website, and if there is then install that too, however, from that point forward I think it will be just fine to install the lasted download version from the support site, including any tools you may want/need and skip the disk, but don't forget your monitor driver and remember every time a video driver gets installed/reinstalled the monitor driver must typically be reinstalled as well.

I install Windows maybe once or twice in a decade so it's not easy to remember how it's done. And then there's sometimes the OS which have been updated in between. 

I have a feeling though, that now when my son is getting older I might need to install Windows a bit more frequently since there's basically two gamers in the house now 

The downloads under Chipset are:

======================================================================================

When the name is the same, user the later posted date of the driver or app. But I can see the question..

A lot of times they will list the actual versions of the drivers or software either on the disk itself or in the disk launched application. If that is so you can compare what is online to what version is on your disk and if there is no update version online, just use the disk itself and then install those items that are a newer version from the online download. 

Make sense?

Get with me before you follow that because of the 10 series and the chipset. I would prefer go over a few things with you first and get a little feedback on them as you follow that outline. 

I know it is balls-on for a 9700K but I would want to confirm some things as you go to make sure it is working correctly as outlined. 

There is more than one way to clock a system and there are other approaches. The approach I outlined is just ducky, although it would probably be considered 'way outside the box' in typical overclocker thinking for the hardware I run and cause some head scratching.

but, but.. that's what I do best  

Did the sticks individually pass in socket A2 with XMP and their correct speed/timing/voltage in the BIOS? and then the entire rack?

I plan to start the PC for the first time later today so I haven't come to memtest yet. 

I'll report back with the result. Each stick in A2 with XMP settings and then all of them at XMP.

I'll check the timings and voltages.

Thanks..  it will be great to hear back the results.

Don't be discouraged if for some reason they 'all' fail..   that may be a BIOS tweak issue for high speed/low latency. If one or two fail and the others don't, that might be a product issue

after loading XMP from the AI overclock menu, reboot and confirm speed/primary timing and voltage, then continue

When you switch sticks or 'rack them up' you must confirm the change in the BIOS first and then continue.

let me know!

You only need to pass 3 full rounds of Memtest86+ per stick and then rack and that can be changed in the Memtest configuration... the default is 4 you can leave it there if you wish.

Your rack is spec'd @ 1.5v, that does not mean 1 stick will XMP that voltage, it also does not mean the rack will XMP that voltage so do look at that and here is my advice based on the high speed/low latency

If 1 stick XMP's a voltage of 1.5..  leave it there.   If 1 stick XMP's a voltage of 1.35, I suggest raising that to 1.45v as that will not be dangerous at all for a single stick DOS test.

It may be possible that each stick tested must be 1.5v so if you get 4 stick fail, try 1.5 per stick at that point. 

That chipset and that CPU are known for being a bit finicky with very high speed low latency memory (above 4000) so keep that in mind as you test and please let me know what happens.

No problem with booting up the first time.

Everything working normally with optimized defaults as far as I can tell.

I haven't updated the BIOS yet because I was curious to memtest a stick with the factory BIOS first

When setting XMP 1 the following settings changed from AUTO to Manual:

BCLK Frequency                     100

PCIE Frequency                      100
DRAM Frequency                    4000 MHz

DRAM Voltage                        1.5V

DRAM CAS# LATENCY             15

DRAM RAS# to CAS# Delay     16

DRAM RAS# ACT Time             36

No problem saving and exit and coming back to the BIOS.

The settings seem to hold.

But these two buddys below have taken a little "jump" in voltages...

They are still on AUTO but with the following values:

CPU VCCIO Voltage:              1.408V

CPU System Agent Voltage:   1.504V

Before XMP 1 they were about 0.95V and 1.05V if I recall.

So a dramatic increase here. 

Would it be appropriate to memtest with those voltages or should I try manual voltages on those two?

I had a very similar experience when I built mine and I mention that in the clocking outline. I was like..  'I'm not even clocked.. are you serious???' 

At this time the BIOS will most likely fluctuate those voltage even though the AUTO readouts are high and yes they need to come down at some point, but for right now I would suggest that you get past the 'individual stick' and 'rack' Memtest86+ tests to confirm you do not have a defective memory module or squirrely memory controller. 

Right now you want to stay with the default BIOS setup until the initial tests, Windows load and other items are checked off and everything is looking just ducky.

But at one point and even if I was not going to clock, I would get those 2 under far better control. I would probably start if HT is enabled with a 1.21v for each but at that point I would also need to run tests to make sure VCCSA is not too low or see where it can come down to. 

We can cross that bridge later, but you are right for bringing this one up..  I was just as shocked as you when I first saw that. 

XMP..  always test the memory at its manufacture rated speed and timing.

and do remember that the SPD is setup for 4 sticks with a (32GB) memory 'kit' so that can influence the AUTO system to run things a bit on the hot side too.

AUTO settings are now:

This looks far more reasonable. Preceeding with memtest now.

I would consider those maxed out and still need a tune at some point but yes the readout makes more sense now. 

I took a picture of the screen before updating, and the factory BIOS was 0222.

1.3v is the Intel specification for both. I put a personal limit of 1.25 on both. Typically VCCIO can live in the 1.0-1.1 range and VCCSA in the 1.12-1.2 range all depending on the individual system and clock. 

The reason is in order to clock there should be a reasonable window of stability with the memory to work with. High tuning memory and then trying to clock can make the process frustrating. If it is possible to trim memory, that should be considered a perk only. 

If a system is not going to be clocked or will not be highly clocked, then working with the presets early on is OK, but just like any other change like that, you must torture test that change just like any other CPU clock change, so best to leave that system alone till later. 

 

DDR4 does run hotter than memory of the past. I can see temps of up to 48-50c under extreme stress @ 1.45v. I do not see a need for a fan for that but if it was higher than that I would probably opt for one. 

Looking around is fun! 

I have huge side door fan too. Just like the NVMe drive that is buried at the bottom of my board and partially behind a audio DAC card, it was running in the 60's under benchmark load and required a small 2.5" pancake fan attached with double side 3m tape which keeps is 48c max under stress and 36c when not stressed. It doesn't take much air. A little does a lot.

The memory modules may need a little help that way too. The large fan may get the air in, but it does not focus the air in specific locations where it may be needed. 

When you get the software installed for your NVMe that is one the items you will need to check with the tower door closed. 3 hard benchmark runs in the software and observe the temp of the drive. If it is hitting above 60c it should have focused air over the heatsink on the board. 

 
Just for fun, I am doing a quick and dirty test checking the benefits of it on my Sandy-Bridge E rig. Using OCCT Large Data Set, it drops the VRM temperatures of about 7-8C, and that's with XMP setting and HT off, CPU at 3800 MHz. So basically stock turbo speed.

EDIT: The Haswell Sabertooth board I ran had the VRM in a duct along with other components that was supported with two 1 inch fans at each end so air was channeled through that system. Was a very well designed board. 

Basically, there's gonna be one stick per day going through memtest here 

But on friday I have some time off and then I will probably be done with memtest.

They corrected the readout flub in the BIOS update so no worries about what you saw initially. 

I thought I had a golden chip with that 100+ reading, now I'm back down to earth seeing 63 

Continue your memtest86 journey. Once you know each stick is good and then the entire rack is stable with Memtest86, it's onward and upward from there. 

Now running stick 3 with the same settings.

It would have been fine at default for this and a few more stages. Just don't clock at anything above those 2 

CPU VCCIO Voltage:              1.312V 

CPU System Agent Voltage:   1.392V   

Apperantly, with four sticks, SA increased from 1.296 to 1.392 while VCCIO remains the same as with one stick.

At this point reducing the speed is an option and since the timing on the memory is low to begin with it should not have much affect on the performance but it would still be nice to find out why. The other option would be to manually raise the CL timing to 16 from 15, leave the speed @ 4000 to see where that sits. 

As long as you know the sticks are stable you could continue on to load Windows. 

With 32GB it may be tough to run 4000 on this CPU and chipset at tight timing and get a clock going too and because of the amount of memory the system may need to run a higher SA than would typically be seen. 

I do not have the hardware to work with myself so my knowledge of the behavior is limited to reading trial and error reports and reviewing what I can. 

Once Windows is installed and since there is a issue running default speed it would probably be a good idea to run HCi Memtest Pro in Windows for at least 500% as a backup to be sure what is being run is indeed stable

Is there another XMP profile you can choose from in the BIOS or only one? If there is more than one is the 2nd a lower speed and what is the timing for that if it is available?