CMD16GX4M4C3200C15 not stable Realbench

CMD16GX4M4C3200C15

I'm using the above 4x4gb Corsair Dominator Platinum Kit in an Asus Z170-A motherboard with a 6700K.

When I built the PC last year I quickly found that XMP wouldn't work, overclock would fail and PC would bootloop IIRC. I found the thread from Corsair Henry saying its the motherboards fault if High Performance Ram doesn't reach its potential. He links to posts from an Asus rep on OCN saying they don't expect people to be buying expensive High Performance Ram for $170 motherboards.

Ok thats fair enough. I didn't realise I was overspending on Ram for the mobo I had chosen. Mea Culpa.

So I didnt use XMP and let the UEFI Asus AI Tweaker Overclock from a preset profile I assume. 4.6ghz for the CPU and 2900mhz for the Ram. I'd see voltage max occasionaly to about 1.4v IIRC. I never noted the Ram voltage. This has run stable for my personal usage for the last year. My CPU 100% Load temps with my H110iGT AIO cooler were around 65º-70ºc IIRC.

I never did any stress testing other than GPU benchmarks and Cinebench15. Never had any crashes or weird behaviour (more on this later) even playing demanding VR games on my Oculus Rift.

So anyway, last week I did some casemods and also delidded my 6700K and replaced the TIM with CLU liquid metal. My Cinebench 100% load temps dropped to a max of 55ºc. The CLU had given me the expected 15ºc+ temp drop. I decided to try and get a better overclock as a result. This time I decided to use the Asus AiSuite3 and Dual Intelligent Processors 5 software to auto overclock. I'd set a max temp target of 65ºc. It gave me 4.8ghz at 1.39v. Great I thought. Hopwever the OC guides suggested ROG Realbench to stress test the Overclock. I used the stress test (as opposed to benchmark) and told it to use 16gb of memory because thats what I have installed. It caused the PC to crash within seconds. Tried again. Same result. I then rebooted to BIOS and reset defaults. CPU was back to 4.2ghz but Ram stayed at 2900mhz. Tried the realbench stress test again. Again the PC crashed within seconds. Went back into the bios and manually set the ram speed to 2133mhz. Ran realbench again and this time it ran for about 10 minutes before......the PC crashed again. Windows logs show error 41 for all these crashes. Given the crashes happened even after I put the CPU back to stock speeds but laster longer when I put the Ram down to 2133mhz, it looks to me like its My Ram and not my CPU thats unstable. While in the Bios I noticed that the Ram is running at 1.2v I think. Should I set the ram to 1.35v??

In the end, I was about to say screw it. Let the Bios Auto OC via a profile to 4.6ghz 2900mhz like last year and forget about instability in a stress test, its stable for my normal usage. Then I remembered that other people with Head Tracking Glitches with their Oculus Rift said that Turning off XMP profiles on their Ram solved their tracking glitches. I had assumed this advice wasn't the cause of my tracking glitches and didn't apply to me as I wasn't running XMP anyway. However, now that I think of it, its said that VR in general and Oculus' camera based tracking system demand the max from the supposed hardware specs. In terms of USB the Oculus hardware demands that a motherboards USB Controllers run to spec. This has caused a lot of people headaches because nothing till now has demaned that much of motherboard USB controllers and mobo manufactueres have gotten cheap and lazy with their controllers not necessarily meating full USB spec's. Some Oculus owners have had to change mobo's or buy add-in PCIe USB 3.0 controller cards as a result. XMP causing tracking glitches for some might also indicate that Ram latency or timing deficiencies etc can cause problems for camera based VR tracking.

So given that my Ram might actually be the cause of my VR tracking glitches however unlikely that once seemed, it also means I can't just revert back to that old 4.6ghz auto overclock even if its not causing crashes for my personal usage because it might infact be the cause of all my VR tracking woes.

Short of buying new Ram, what can I try, do, adjust to try and get my PC to pass a realbench stress test even at 2133mhz. Remember Real bench is causing a crash after 10 minutes even with the CPU at Stock and the Ram back at 2133mhz!! I assume if it can pass that, I can rule it out as the cause of my VR tracking glitches.

Sorry for the wall of text! :o:

TLDR:

My 6700K running at stock clocks and my Corsair 3200mhz DomPlat running at 2133mhz will crash after 5-10 minutes of a Realbench Stress Test.

What can I do??

Do you have any other control programs like MSI Afterburner or EVGA precision running at the same time? This has been a long standing issue and there was a supposed fix, but I wouldn't bet on it.

I have that same kit and although I run it on an X99, I am surprised it would have difficulty at 3200 and certainly at JEDEC 2133. I am not a big fan of the AI Suite overclock and the flaws are fairly apparent in the way it is set. You can still use it for a base idea, but even if you keep the CPU settings, I would set the DRAM values manually in the BIOS after you have finished. I don't see a lot of reason to run the non-standard timings and the results are very inconsistent.

I was posting in parallel in the Cooling forum when I read monitoring programmes conflicting might be the cause and asked a question about turing SIV off for the stress tests. Turns out I wasn't getting crashes. It was only when I thought to unplug my monitor and plug it back in that I saw the PC hadn't crashed at all. Luxmark part of the test was failing. ie. It was GPU driver failures. I disabled Thundermaster, Palits control program. Stress test lasted 4 minutes this time but the same 'crashes' occured.

Anyway, it looks like my ram might be OK after all. So I'll try overclocking again but use different stress tests this time. Well my Ram is OK in so far as it'll overclock in this Asus Z170-a board. Didn't realise I was overspending on Ram for the 'cheap' Z170 board I bought, but Corsair Henry's sticky here and an Asus Rep on OCN bascially say not to expect rated ram speeds on lesser boards.

So when I use DIP5 to overclock again, I should go into the bios and adjust the timings? Don't mess with Voltage for the Dram?? On a sidenote, In DIP5 I was hoping to set a Voltage target of 1.35v for the CPU, like most of the guides say, but for some strange reason my voltage target slider doesn't not have a volt or millivolt number like the screenshots, it has a value range of + to +125. Weird!! I can still set a temp target though. Can you suggest a temp target for the auto OC given I'm runnning a 6700K delidded with CLU and a H110iGT. When I was running at an asus TPU2 switch Auto OC of 4.6ghz I was hitting about 55ºc with 100% load running Cinebench15.

My experience in testing the various auto overclock methods with DIP5 has always been inconsistent. If very often ignores the parameters I set and will "visually" overclock into oblivion. On X99 boards and CPUs, the voltage table it uses makes little sense and it often refuses to set the frequency to the target level, even after passing the stress test at that level. I have been running for a year at 4.5@1.275. It thinks that is a better voltage for 4.3Ghz and refuses to set 4.4 with it or anything higher. I really think you can do better with an educated guess and certainly with a little testing. The "stress test" built into my version barely challenges the CPU or my cooling. Peak temps are usually in the mid 50's when running that. That makes any kind of temp target a bit meaningless. Delidded, I suspect you will have the same problem. Stability limits will precede temp limits, unless you use absurd voltages. You are likely to hit the safe voltage cap before the temp limit. I am not sure where people set that limit for 6700K, but it is probably around 1.40v for daily use.

As for the memory, no voltage boost needed. In the AI Tweaker column, set the DRAM frequency to your target (3200). Set DRAM voltage to 1.35v. Then enter the DRAM timing voltage sub-menu. Enter the primary timings of 15-17-17-35. (might just be 15-17-35 on Z170 boards). You could also enable XMP tuning after doing your AI Suite OC, but I think it complicates the issue and Asus boards seem to better when allowed to follow Asus secondary and tertiary timings. All of those will be on Auto. If you need or want to try a lower frequency, you should be able to run 2666 (13-15-15-31)@1.25-1.30v. I can run this kit at 12-13-13-28-T1 at 1.30v when at 2666MHz.