Problems with DDR4 3600

deathstalker616 · May 4, 2017

Hello there. So I have recently built a new rig, but have been running into some problems. My rig is a i7 7700k in a Asus Maximus Formula IX. My RAM is 16GB (x4, total of 64GB) with 3600 MHZ speeds. I have a GTX Titan X Pascal hooked up to two Asus ROG Swift PG278Q. For the most part, my system is working quite well. However, I have an issue on boot whereby the computer will run but not POST, nor will anything be displayed on the monitors. This situation happens entirely at random, I could restart 5 times and assume that it is working fine, only to have the issue reoccur later. Generally it tended to happen within a few restarts. The only way I found to solve the issue would be to reboot my computer, whereby I will be told that it failed to POST and I need to enter the BIOS. Even if I don't edit anything, my computer will usually load after that. I should also note my computer works perfectly fine if it gets past this boot and whatever it is doing is not severe enough to corrupt installations, I can do all of this in the middle of an installation and get functional software. I assumed the issue to be related to something display related, like the monitors or graphics card, but after playing around in the BIOS and restarting a total of at least 95 times (that I tallied) I feel confident enough to say I have found the source of the issue to be my RAM.

The solution is simply to drop it back to 3466. I turned off XMP and did a manual load to 3466 with the same timings and voltage as 3600. This instantly fixes my problems. Yet I have the timings set to 17-19-19-39 at 1.35V, as it should be advertised for what is required by 3600. So my question is, what do I need to do to get 3600? I am assuming perhaps a voltage increase, but I am inexperienced with doing such, I mostly have just relied on XMP in the past. It also seems illogical to go beyond what was specifically tested for my particular RAM. I almost wonder if it is something to do with the BIOS having trouble seeing this recent type of RAM. I have the latest version, 0906, which was made in March and I assume would have corrected the problem if it existed. It could be a possibility, so has anyone had success with this particular model motherboard/RAM combo?

I also am not sure why, but it seems setting it to 3600 causes anomalies to happen in general, especially if I do it manually without XMP. I got an error about my keyboard not being detectable on one test. Another resulted in a blue screen, saying my OS was corrupted (it wasn't). One thing that seems consistent is that the sidebar in the UEFI displays 2133, the default, when I am actually able to get it visible. I should note that exiting the BIOS without making any changes, I have made it to the desktop with the correct speed as stated by CPU-Z.

A bit of another question, but the sidebar I mentioned also shows my 7700k at 4200MHZ, which is odd because the target is 4500MHZ according to the UEFI and the turbo is enabled. According to CPU-Z, the turbo seems to be enabled. Yet the XMP changes to whatever is the current setup, so I wonder if this is stating that it is not accurate.

MeanMachine41 · May 4, 2017

You are using 3600Mhz OC RAM and the i7-7700K CPU officially supports DDR4-2133/2400Mhz.
As OC RAM @ 3600MHz is not a JDEC standard then your XMP profile would not be recognized.

64GB of 3600Mhz RAM is quite a bit to ask of the IMC on the chip and the probable cause of your instability and Post loops.

With OC RAM your XMP profile may not work without OCing the CPU and then there is no guarantee.

As you have the latest Bios 0906, then you will have to rely on your Bios Tweaking expertise with Primary Timings and Voltage to get them to work.
Either settle for 3466MHz (very little difference in performance) or try a moderate OC on the CPU and see if you can then stabilize your RAM. Also try with just 16GB of your RAM installed if your kit combo will allow.

deathstalker616 · May 5, 2017

I see, that would explain why I have to do everything manually.

Yes, I have inferred that much through deactivating the XMP and slowing up the RAM. It is quite the demanding thing to ask, but I assumed since it existed that it must be supported by recent high end motherboards/CPUs.

I have taken the solution about trying to overclock the CPU, using the first overclocking profile as a guide I have gotten 4.8MHZ (48-48-47-46) stable and can probably take it farther. I figured with such a small amount of MHZ difference that it would allow for 3600 RAM, but it still seems to sadly not work. I should mention I have the Load Line Calibration on Auto, setting it to 3 as the profile suggested caused an instant failure. Similarly I could not change my Current Capability from Auto to 140% without post loops on some restarts. Not sure if either would affect my RAM stability, but it doesn't seem to have any change in instability of my CPU.

I am quite inexperienced with Bios Tweaking, but I am playing with settings and hoping to see what I can achieve. I shall see if I can go for a bit of a higher clock and if that solves the problem. 3466Mhz is very minimal of a difference and I am prepared to be happy with it due to the minimal difference, but I did chose it over 3466 RAM, meaning I would like to get the speed I paid for if possible. I am not sure what limiting my RAM to 16GB would do, since it is stable at 3466 on 64GB, am I testing the 3600 speed to see how it works on a singular RAM chip?

MeanMachine41 · May 5, 2017

It would seem your IMC (Integrated Memory Controller) is not up to the task with default voltages.

Sometimes by using less RAM or a single module they work. (Less strain on the IMC) in which case you can try increasing the DRAM Voltage (VDIMM) or VTT (goes by different names depending on the MB) which will help stabilize. Do not exceed 1.4V MAX for Intel CPUs.

Go to Extreme Tweaker menu>AI Overclock Tuner and set to "Manual"
Check your TPU and what selections you have. They will differ depending on your cooling system.
Disable CPU Cstates. This is just a power saving feature and can spoil a decent OC.
LLC set to high or 5 will help with voltage droop as you increase your CPU frequency.
You will have to stress test the system at each phase for stability.

deathstalker616 · May 5, 2017

I have successfully gotten to a 4.9MHZ stable clock, it seems 5.0 causes blue screens for me after heavy usage. Still not working with 3600, so I don't think an overclock is the solution I was hoping for. Regardless, it is nice to set up, so I will be keeping it at 4.9MHZ.

I am using a Noctua NH-D15 placed right above the G.Skill TridentZ RAM. Needless to say, I have RAM with long fins right under the CPU cooler, making it not the most convenient of things to remove. Regardless I have increased the DRAM voltage from what I have been using (1.35) to 1.39, didn't seem to make a difference.

AI Overclock Tuner has now been set to manual. Not sure if this changed anything since the Base Clock is still the default value.
My TPU is currently set to respect all current settings. I could change it to TPU I or II, 1 being air and 2 being water. As I have a Noctua NH-D15 I went with TPU 1, it caused what has been referred to as a post loop (still not sure if that terminology is exactly right, this not looping but instead permanently stalling).
CPU Cstates has been disabled. Glad to be made aware of that one.
I had LLC set to 3 (Apparently LLC doesn't work with a basic turbo, but works fine when you have a high overclock), but I have changed it to 5 with no changes.

All of this has been stress tested. I use a Cinemark CPU test for a quick run and a Adobe Premiere render of a 5 minute video to see if it can sustain itself for a few minutes. Longer stress tests will probably be necessary, but I can imagine it to be relatively close if it doesn't happen within the first few minutes.

Edit: Actually, with the recent changes suggested, I was able to push my CPU up to 5.0. I am a bit cautious due to the previous bluescreen, but I believe I have hit a 5.0MHZ clock. It just still dislikes the 3600 RAM sadly.

MeanMachine41 · May 5, 2017

Well a 5.0GHz OC is very good if stable.

Download AIDA64 and HWInfo monitor and put them on your desktop together.
In AIDA64 go to Tools and select the stress test. Choose to stress your CPU, FPU and Cache.
Run for 10mins and keep an eye on your temps. Stop the test if temps reach 75C.
Take screen shots at the 10min mark and post your results.
From this stress test you can determine stability of voltages and temperatures under load.

I think your temps will be too high using AIR cooling however the test will tell.
Yes it's a nuisance when the cooler blocks your ability to swap RAM however 64gb @ 3600Mhz RAM may be just too much for the IMC.

deathstalker616 · May 6, 2017

Hmm, well, this isn't promising. According to the tests, even 4.2GHZ runs too hot with a couple spikes jumping into the 80s. I believed the cutoff point was the 90s, plus the Intel Extreme Tuning showed nowhere near these levels even when running a Cinebench. But, basing it on this, I might have a lot of problems with stability I am unaware of. For some reason it seems to be that my cooler isn't working at all, when the fans are operational. This definitely makes things more concerning. Note these issues are popping up within the first minute.

Yeah, it seems like it will be not likely that I will be able to use 3600MHZ RAM sadly. Of course now I am wondering how stable this computer is at all, this test shows information in a lot worse of a light than I previously believed.

Karadjgne · May 6, 2017

Hopefully you are resetting default values between these different OC attempts. The settings you had at 4.9GHz will be different than those at 4.2GHz, and if you happened to have missed one you manually set, it can throw everything out of whack. This is OC, and it takes patience and attention to steps taken, starting small and working up, it's not just a matter of jumping high after a few settings changed. Adjusting current capabilities to 140% is almost never needed, my 4.9GHz is set to high (50%) LLC and 110% CC, cpu/vrm spread spectrum disabled, PLL at 1.7 instead of usually going up to 1.9v etc. But all that took time. I got stable at 4.8 and couldn't go higher, because everywhere I read said I'd need to increase PLL to 1.9v to compensate. Moved it down from there to 1.7v and moved to 4.9GHz stable. At that speed, 1 little change can mean a world of difference in either being or not being stable, regardless of actual vcore voltage. I even use a negative offset mode, not the more common positive. It could simply be you haven't adjusted your long/short durations or primary/secondary plane current limits to accommodate higher OC limits.

So what I'm saying is that there definitely is hope, but your fast answers belie the opportunity to really study your OC and test what works and what doesn't.

Oh, forgot to mention, disable any and all EPU power saving abilities, they'll choke an OC faster than your wife can change her mind.

MeanMachine41 · May 6, 2017

deathstalker616 :

Well yes it's when the system is under load when it counts and AIDA64 will certainly test your system.
Air cooling is not the way to go if you want to OC.

deathstalker616 · May 6, 2017

Yeah, Karadjgne, I mostly was just changing the multipliers and seeing how it affected the temperatures. I was trying to do things at a relatively slow pace, but I guess I was moving too fast without realizing it. I should definitely go back to basics, this definitely isn't going well otherwise. As such I have decided to run a basic test just to see how the temperatures handled. EPU power saving was actually disabled by default, thankfully.

I can definitely see what you mean MeanMachine41, though I must admit I am a bit confused. Air cooling generally is not the best, but a Noctua NH-D15 tends to hold up extremely well alongside liquid cooling. I had actually heard it was possible to get 5GHZ stable clocks with a Noctua, but perhaps I am mistaken somewhere. At the least I would expect it to perform better than a stock cooler. Deciding it best to perform a default test to see how it functions, I have gotten this - https://ibb.co/j5rhFk

I did disable C-States (default being Auto) as was suggested previously due to it helping stability. I then ran another test, which pretty much shows it didn't do anything bad to the temperatures due to them being within the same range as they were previously. Regardless, this also shows that the temperatures I am getting are consistent. Core #4 in particular tends to spike up. - https://ibb.co/d6h8MQ

As you can see, generally speaking the temperatures are quite nice, but as stated they can spike upwards. I am nearly hitting the 75 limit on Core #4 without even changing a value from the defaults.

Karadjgne · May 6, 2017

Adjusting c-states won't affect stability or temps during normal running usage. OC, in part, means speeding up the pc but that raises temps, so to combat that, voltages and current are lowered to as minimum needed to maintain stability. That's at c-states 0. When your pc goes idle, sleep, it enters c-states 1+, at which time the cpu starts lowering voltage and current usage, stepping further down with every c-state. It's this that affects stability the most, where it's entirely probable that the cpu will step down too much, and with limitations set in bios by you may be beyond the ability of the cpu to recover, not having sufficient power flow. When that happens pc crashes, bluescreens, refusal to wake etc. Older gen, such as Ivy and Sandy-Bridge had higher power expectations, so this wasn't so much an issue, but with Haswells and newer, c-states are ultra low power settings, so this is an issue. Same goes for Intel speed step, to get the absolute best performance out of a pc, the very highest possible stable speeds, the pc needs to be straight line, no power saving at all, so speedstep should be disabled, as it's a power saving setting, reducing cpu to 1600MHz cores, but reducing power needs too, maybe beyond recovery. Stable OC requires stability in everything, that includes stable voltage (job of LLC) to reduce fluctuations in power. A decent psu can help, or hurt, necessitating maybe the outputs of the very best psus to get DC voltages as tight as possible, straight line.
There's many things that can affect power fluctuations, figuring out exactly what does and to what degree is just part of extreme OC.

Old school cpus hit thermal limits long before voltage limits, so rose the need for the biggest coolers, both air and liquid. With skylake and newer architecture you are more likely to hit voltage limits than thermal, so the upper wattage heat output ability of the big liquid coolers is very rarely used, that Noctua NH-D15 is quite capable of handling a 7700k at max. Intel, however, doesn't have the tightest reign on its own TIM application between the cores and lid, so core temp variations do happen ±5°C is considered normal. So core #3 can easily be 10°C above core #0 etc. And that's assuming perfect application of user TIM between cooler and lid. While I don't endorse the solution, many have used the delidding option (done correctly) to get better temp control and less core temp variations.

deathstalker616 · May 6, 2017

I see, power saving definitely is not a beneficial thing to have, especially when using a desktop rig. I have a AX1500i power supply, so I do not believe I would have too little power. Regardless, I don't think I can increase my clock speed due to the necessity for more heat generated, I am essentially at the edge with stock speeds for a Noctua NH-D15 (which I don't understand unless the Thermal Compound was applied incorrectly). I shall deactivate Intel Speed Step if it isn't already deactivated to ensure better stability. I guess this is my biggest confusion, a Noctua NH-D15 should allow for at least a mild overclock, but what I am seeing looks like stable is just about functional. Am I concerned about this 10°C difference for the limit of 75°C or am I just needing to ensure the CPU itself doesn't hit that speed? I have hit 77 on some tests on Core #4, which makes the situation even more concerning.

I am aware that the TIM on the 7700k is not exactly the best, but I didn't think it was this drastic. Regardless, the delidding process looked complicated based on what I saw in the past, I was concerned about damaging my CPU and therefore decided it wasn't worth the hassle. Plus I needed to be sure the computer worked in case I needed to return some parts under warranty.

Karadjgne · May 6, 2017

The D15 should be more than enough to handle a 7700k at a very good OC. You definitely have a thermal issue somewhere, I'd start by re-evaluating the cooler installation and paste application. If that solves nothing, I'd reflash the bios with the newest update and try that at stock. Leave the ram at 2400, push the OC (correctly this time) as high as you need/can get, then start bumping up ram speeds. You'll probably find that you won't need to change the OC at all to set 3600 via xmp because the cpu (MC) is now better equipped to handle the 4x dimms. You may need to bump vram voltage up to 1.355 or so.

deathstalker616 · May 7, 2017

Hmm, I might need a bit more of an explanation of what I am evaluating. The aftermarket cooler was the part of my computer that intimidated me the most, before this one I had only used stock coolers. It appeared the internet had very conflicting ideas on paste application in particular, every individual seems to believe it should be a different method. Part of one of my co-workers jobs is handling computers, so I decided to pay him and watch the application than take a risk myself. I would hope he set it up correctly. He put a relatively decent sized blob on the CPU and spread it out with a piece of cardboard to cover the entire surface. I have heard about there being a pea method as well, but his argument was that you want there to be no air holes between the CPU and cooler. Even if your personal method is different, is this method fine or is there something detrimental I should be aware of?

Karadjgne · May 7, 2017

Intel cores are not square. They are a strip going from top to bottom, covers about ⅓ the die, kinda like the white part in the French flag. This is the most vital area to cover correctly. The corners and very edges are almost unimportant comparatively. However, the actual surfaces of the cooler base and lid are not perfectly even. The job of the paste is to fill in any micro scratches, abrasions, the concave/convex nature of the meeting of those 2 surfaces. Again, however, paste is not a very good thermal conductor, but immensely better than air. So it needs to be thin as possible, while still spread to cover the majority of the lid. Paste too thick is just as useless as paste too thin, as it insulates the lid, not allowing the heat through very well.

So, application isn't all that important either, ppl use the grain of rice, small English pea, X, credit card etc and they are all pretty much the same if done correctly. I use grain of rice(ish) rather than a round pea, works fine. You can test size amounts by pressing the cooler base down on a glass top etc, it'll be pretty small. Just don't use too much or the paste is too thick, spreads too far and ends up in the socket. You actually have a greater chance of air pockets doing the cardboard thing if there's a dip in the paste. The only thing to make sure of is clean surfaces, all traces of old paste, lint, dust etc need to be removed. Best to use a coffee filter (several) and 91% or higher isopropyl alcohol.

Paste is important too, you want one that spreads easily, isn't thick or pastey, and requires minimal effort. Gelid extreme 3 or Noctua NT-1 is about the best for this, stay away from arctic silver 5, its very mediocre and has its own set of issues. Properly applied, they'll easily last 10 years or better, AS5 really needs a redo every year or so. It's honestly not much better than toothpaste.

http://www.tomshardware.co.uk/thermal-paste-heat-sink-heat-spreader,review-32799.html

A good read on this subject.

deathstalker616 · May 7, 2017

I am appreciative for all the details, I have attempted to research various information in the past but often find it a bit confusing. Your explanation is quite easy to follow.

The biggest concern I have about removing the air cooler at this point is potentially wasting an application. If I remove the air cooler, I assume I will need to reapply the solution even if it was properly done the first time, which would be made harder by having to remove all the old paste. Unless it is absolutely necessary, I am not sure it is the best of ideas to waste thermal paste.

I had heard many praise Arctic Silver 5, but found NT-1 to give slightly better temperatures and be a better paste overall, I had heard in the past it was one of the best you could get. It was included with the NH-D15 and was the paste I ensured was used.

Karadjgne · May 7, 2017

The only reason AS5 became as popular as it did is because back in the day, paste was relatively little understood and not really thought of as all that important. Retailers were shipping very small tubes that cost $20 or more. Then Arctic started selling the AS5 in tubes 3x the size, that cost just $6. It took off like wildfire because you now had plenty, even if you screwed it up several times. And it worked. Ppl didn't care if they had to repaste more often, it was cheap and they had plenty to spare. And at the time, probably was better than most. But nowadays, heat is much more understood and is much more vital to enthusiasts pc's applications, so correct paste is vitally more important.

That D15 is well capable of handling anything an i7-7700k can dish out, if the paste is correct. If it's too thick, you'll see much higher temps as it gets harder for the lid to transfer heat to the heatsink, it's basically insulated.
At stock settings with that cooler, you should easily be able to see (at room temps in the low 20°C's) idle temps in the very low 30's and aida64 temps in the mid 60's. Gaming should be in the upper 40's to low 50's at best.
With a high OC like 4.9GHz, even aida64 tests shouldn't be climbing much if any past 80°C. My i7 at 4.9GHz on a kraken x61 sees 74° with p95 26.6 small fft at a 23°C ambient, and at those wattage ranges, the D15 is equitable.

I personally think your paste is simply applied a little too thick, or the cooler isn't tightened down enough on the cpu lid and you aren't getting optimal heat transfer, so seeing higher temps under load, especially when the MC is stressed with 4x sticks at 3600MHz

deathstalker616 · May 8, 2017

I see, I do recall there being a point where people would say that toothpaste could work in a pinch and thermal paste is only slightly better. I guess I didn't realize that the NH-1 was that significant of an improvement, glad I decided on it.

So I removed the cooler today and was in for quite the shock. It turned out the technician had applied a massive amount of thermal paste. Removing the cooler, which I assume would be similar to the glass test mentioned earlier, resulted in a giant glob being pulled with it. I spent a bit of time cleaning it up and ensuring there was no remaining thermal paste before applying a new dosage.

I used a cross method, keeping it relatively thin but with the ability to expand when the cooler was pressed down. I ensured the CPU cooler was correctly tightened and the fans were facing the correct direction. These were the results - https://ibb.co/gFTBRQ

So is this still a bit too hot? It seems Core #4 is the only one that is going past the mid 60s for the maximum temperatures, while the minimums seem about right.

Karadjgne · May 8, 2017

67,68,68,72 is all within normal ranges, there's just 5° difference in min an max, which is actually pretty good as it can be anywhere upto @10° difference. There'll always be the odd core for some reason run slightly hotter.

So with those new results, what's the test at? Full stock with 2400? Or is that some OC and 3600?

deathstalker616 · May 8, 2017

That is promising to hear. I was concerned by the apparent lack of change on the thermal paste, especially since some tests prior to changing the CPU such as the ones posted here were no higher than 72 but after a few trials went to 77, though I have yet to run a test that passes 72 where previously Core #4 would often spike up to 77. Much like how 77 became a constant max speed the more I tested it, 72 seems to be remaining as a constant max temperature for Core #4.

Mostly stock. the speeds are still the same at 4.5GHZ with turbo, but with C-States disabled. Changing this has done nothing for the RAM speed allotment sadly, so my RAM is running at 3466. 3600 suffers from either POST loops or errors about my keyboard not working/my SSD not existing. For some reason it seems more unstable to do it manually than set it via XMP, though both are relatively unstable and unfit for normal operation. Timings at 17-19-19-39. DRAM Voltage at 1.35.

Karadjgne · May 8, 2017

17-19-19-38 - 1T is decent timings for 3733MHz ram, with 17-18-18-38 - 1T for 3600 and 16-18-18-38 - 1T for 3466, so your timings aren't bad, but could be better. With 4x dimms, having slightly relaxed timings isn't a bad thing. You could try manual adjustments, maybe 18-19-19-38 and bump voltage to 1.36v.

It's funny, I can push my 1600 to 2400 quite easily, its actually at 2133 10-11-10-30 1.515v right now, and you are having issues getting to rated speed, never mind any OC. So much for technological advancement.

deathstalker616 · May 9, 2017

I figured I could get away with better timings, I hadn't tried to do such due to the focus being on 3600 RAM, thus I had kept it on the timings stated to work with my particular 3600 set. I was able to set the adjustment to 18-19-19-38 without any problems. No temperature changes worth noting. Stability is lasting at least as long as the AIDA test. I should mention I did not change the voltage from 1.35v, I assume I do not need to increase it unless I find myself with unstable timings. That is the good news about this 3600 RAM, it works quite well at 3466 apparently.

Indeed, it definitely is an odd issue, though I assume it is a lot harder when you hit the upper threshold of memory speeds. 3600 is the fastest speed you can purchase for 64GB RAM, so I guess I am just struggling with the upper ceiling. It appears I just need to settle for marginally slower speeds, which perhaps balances out with the lower CAS. With decent temperatures and still pretty fast RAM, it is hard to complain about the setup.

Karadjgne · May 9, 2017

Very true.

Problems with DDR4 3600

Prominent

Judicious

Prominent

Judicious

Prominent

Judicious

Prominent

Titan

Judicious

Prominent

Titan

Prominent

Titan

Prominent

Titan

Prominent

Titan

Prominent

Titan

Prominent

Titan

Prominent

Titan

Share this page