I need help stabilizing a CPU overclock

SickBeast wrote: ↑Hi Guys,

I have an Intel i5 4690k CPU that has been overclocked to 4.5ghz for a long time now. Recently I upgraded to 32gb of DDR3 2400mhz RAM and that's when my instability problems started. I did some reading and apparently Haswell can become unstable when you overclock the CPU and the memory at the same time.

What's really strange is that my computer passes all the stability tests that I run (Aida64, Prime95, Intel Burn Test, OCCT, ROG RealBench, plus 12 hours of Memtest86+). What's happening is the computer will give me a BSOD about once a week. Sometimes the BSOD says "memory_management" and when I look in Event Viewer it's kernel related.

If any of you guys have any tips on how to stabilize Haswell I would really appreciate it. I have tried giving more voltage to the ring bus, more system agent voltage, plus a few others. I went through every possible option in the BIOS and tried to give it a little bit more juice but my system is still a tiny bit unstable.

Right now I have the CPU at 4.4ghz at 1.2v. I was running at 4.5ghz at 1.25v for months and I never had a BSOD until I upgraded the RAM. I am getting BSODs with the RAM at stock and also when it's overclocked to 2400mhz. The RAM passes Memtest though at both settings.

Here are my system specs if that helps:

Gigabyte Z97MX Gaming 5 (latest bios)
Intel Core i5 4690k
32gb G.Skill DDR3 2133 RAM @ 2400mhz (1.65v, standard timings)
Thermaltake 750w PSU
Sapphire Radeon 280x

Thanks guys I hope you have some ideas for me!

Try setting the RAM modules to run at 1600 speeds and the errors should clear up, I have seen through the grapevine also that the rated voltages above 1.5v can cause issues with the CPU.

whats your system agent voltage at now?

youve added more volts to RAM,sysagent, CPU all at the same time and it still BSOD's?

What about memory input voltage? ring voltage?
What are the timings on the memory?

Usually when you add 32 GB to any board it stresses the memory controller even though the chip is designed for it. Since you overclock it makes that fact even worse.

I would start with loosening the timings, REAL loose see if the problem persists. Do this with a a small increase in ring voltage, memory input voltage, VCCSA voltage. Also bump up your LLC by a notch or two if you have the option.

I am kinda new to Haswell/E. I have an X99 and not sure if the options are the same but my google skills tell me you should have the above options

Hey sickbeast, despite our differences in the past, I think you should test your OC rig a little differently. You need to run prime95 max ram usage with a looping 3d benchmark/demo for 24 hours. What you're seeing is classic behavior when only 1 part of the cpu is being used but another part, like mmx etc isn't. When both are being used then you see the cpu freak out.

You can loosen the memory timings - memory timings and overall memory speeds does practically nothing for frame rates. I rather have a faster overall cpu and slightly slower memory than the reverse.

yeah perfect please do.

if you're certain you've hit all the 24/7 safe limits on all voltages relating to memory controller, memory, SA, ring, LLC... Theres not much else it could be voltage wise. Do you have CPU input voltage as well? I know its not supposed to correlate with RAM but I know on x99, upping that voltage can maintain that extra little bit of stability.

Im assuming your RAM is profile rated @ 1.65v @ 2400??

Do you have a 'memory input voltage' setting? Not memory voltage, its something else.

EDIT: what model is the memory?

Try changing your multiplier and leave your ram at stock clock and timings. If you are overclocking your ram you need to check your timings in the bios. If they were left alone and you overclocked the ram you are going to run into issues. Either overclock your ram and loosen up the timings, leave the ram clock at stock and bump your multiplier up(if your cpu allows) or loosen timings. Might be a combination of those three things or just one of them.

also, if your ram is rated at 1.65V try bumping it up to 1.66V. You have a lot of ram there and the board may need a bit more juice.

He is touching on 80% of the algorithms that you should use to test stability.. I highly doubt it has anything to do with the particular benchmark software hes using.
This is either a power issue; whether it be a bad PSU (unlikely), voltage needing adjustment, bad RAM.. or the highly unlikely event that windows is causing it with a pure coincidence in timing with replacing the new hardware.

rageking wrote: ↑Hey sickbeast, despite our differences in the past, I think you should test your OC rig a little differently. You need to run prime95 max ram usage with a looping 3d benchmark/demo for 24 hours. What you're seeing is classic behavior when only 1 part of the cpu is being used but another part, like mmx etc isn't. When both are being used then you see the cpu freak out.

You can loosen the memory timings - memory timings and overall memory speeds does practically nothing for frame rates. I rather have a faster overall cpu and slightly slower memory than the reverse.

I have the same CPU, and frankly I just left mine at 4.4ghz and called it a day, I couldn't keep 4.5ghz stable if my life depended on it.

However if you're overclocking the ring bus try leaving it at stock or at least not super overclocked. Supposedly it makes very little real-world difference but makes it more unstable to overclock ring bus.

Found this as an 'optimal' guide to OCing memory at those speeds on your particular board.

[IMG]http://i263.photobucket.com/albums/ii15 ... m_info.jpg[/IMG]

hmm yeah those voltages do look on the high side.. and yes, some boards are extremely finnicky with 4 DIMMS if the RAM isnt on the QVL.

anything sandy bridge and above, I usually suggest to people that they try and purchase memory thats on the QVL.

you can check the qualified vendors list on the gigabyte website.


relax timings on the memory. let us know what happens.

I wouldnt add any more voltage

Keep the memory at 1600 or even 1333, and stick with standard JEDEC timings (i.e. 11-11-11 for 1600). Faster memory is useless for consumer workloads if you're not using integrated graphics.

You should save yourself the trouble and just set it to 1600.

SickBeast wrote: ↑I'm going to run some benchmarks when I have some time and I will try to prove or disprove the notion that faster memory makes no difference. I know that in a lot of applications it can give you a nice boost.

What are these applications you speak of? It certainly doesn't affect any games.

SickBeast wrote: ↑Well it turns out that my RAM is also stable when running on its XMP profile at 2133mhz, except I have to add 0.05v extra to make it stable. So it runs stable at 2133mhz at 1.65v. It also runs stable at 2400mhz at 1.70v. Is this normal? Is it because I'm running 4 dimms? I'm thinking about RMAing the RAM if it fails in a single stick configuration at its rated speed and voltage.

I was going to mention this before but I figured the IMC in Haswells is at the point where it can easily handle four DIMMs (unlike Lynnfields--not so much lol). If you look at the RAM support list there's lots of configurations with four DIMMs that are acceptable so I didn't bother saying anything. But, try pulling two DIMMs and see what happens. Then if it's stable, try the same DIMMs on the other channel and see. Still stable? Then pull those two sticks and try the other two you initially took out, on either channel.

XMP is also an interesting thing. There's more settings encoded in XMP than the basic/standard timing set of like 9-9-9-24 2T (or whatever it would be) so it seems when you use XMP things work very well. Sometimes you can turn off XMP, set things identically manually and boom it won't even POST Very odd XMP is, lol. In other words yeah definitely stick with XMP settings. Note that on some RAM with XMP support you'll have two profiles... On the more modern version of XMP one is the stock settings and the other is a "bleeding edge" performance profile that is not guaranteed stable and is typically used by those that are trying to extract the maximum out of the memory.

I have the same motherboard (Z97MX) in one system, incidentally. However I have an i3 in that one so no OCing. The RAM I have in there is four DIMMs though--DDR-2133 HyperX, in 4x2GB configuration (8GB total). I just used the XMP settings though and it runs perfectly fine with the stock speed. This kit I got for cheap some time ago, it was actually designed for quad-channel use (LGA2011) and I initially used it on a Lynnfield system--no dice! Lynnfield CPUs have serious troubles running four DIMMs that fast and I couldn't even get to stock speed, not even close. I think DDR3-1700 was as high as I could get with four in (with two I hit 2133 no problem though), and it took me quite some time to understand that the IMC just couldn't handle it. However as said, I would have though Haswell wouldn't have a problem. But then again you have 32GB, so that's four 8GB DIMMs which may be pushing it even for Haswell. I'm not sure what you're doing that you feel you need 32GB, but you should at least try with two DIMMs and see if that alleviates any problems.

Keep in mind it may also be your board or PSU that's requiring you to go up .05V. The board may not be holding 1.65V under load with that much RAM (and four DIMMs) installed. This is a mATX board we have here. If it's all kosher at 1.65V at 2133, I'd just leave it at that. You probably don't need to push it to 2400 if you need 1.7+, unless you're really not all that concerned about longevity and 100% stability.

Edit: Sorry my "Quad channel" kit is actually DDR3-2400, not 2133! It does work at that speed (I'm pretty sure it's set to that) with the stock-clock i3, and all four DIMMs, with the XMP profile. However this is only 8GB, not the 32GB you're running.

board123 wrote: ↑What are these applications you speak of? It certainly doesn't affect any games.

Oh no?

http://wccftech.com/fallout-4-performan ... to-report/

:P