PDA

View Full Version : XMP Instability



PilotGW
11-24-2017, 05:32 PM
Hi,

For the better part of a week I’ve been battling (and failing) to try and get my 8700k stable. I’m only aiming for 4.8GHz. That’s the maximum I’m comfortable with in relation to temperatures (and I know I lost the silicon lottery anyway so I won’t bother).

My specs are:

8700K
Maximus X Hero
16GB Corsair Vengeance Red LED 3000MHz
MSI 1080Ti
Windows 10 Pro

I currently have the following BIOS settings:

BIOS 0802
MCE Auto
Sync All Cores
Ratio 48
AVX Offset 0
LLC 5
IA AC DC Load Line 0.01
Adaptive Voltage 1.312v
Offset -0.016v

These settings give me a VCore under load of 1.296v.

If I leave the memory at stock, it seems stable. However, if I turn on XMP or manually set the memory overclock things fall apart:


Prime95 26.6 Blend: One error detected within minutes (although I’ve only done this once overnight)
OCCT Linpack: Error detected at exactly the same time, every time (around 10 minutes 45 seconds)
ASUS Realbench: Instability Detected (7z), every time (usually between 20-40 minutes)

I’ve tried playing with the IO and SA voltages (XMP turns IO up to 1.34v and leaves SA at 1.1v!), same result.
I’ve also tried getting rid of the overclock with just XMP enabled, same result.

At the moment I’m thinking dodgy memory. I did run Memtest earlier in the week and after 7 hours threw 2 errors. Subsequent runs of Memtest however have come up clean so I’m not sure.

I’m at my wits end and don’t know where to turn next. Any help would be greatly appreciated!

Hopeywolf
11-24-2017, 05:40 PM
So XMP profile at stock will give errors? I am leaning toward bad memory on this one.

Silent Scone@ROG
11-24-2017, 06:03 PM
You'd need to test the memory at optimised defaults to know if it's bad sticks. If errors are only occurring when XMP is enabled, then it's memory instability.

I don't recall if the Maximus X has Tweak Modes under Tweakers Paradise/Extreme Tweak. If it does, try setting Tweak Mode 1 and retest the memory.

PilotGW
11-24-2017, 06:07 PM
So XMP profile at stock will give errors? I am leaning toward bad memory on this one.

Yeah, that's what I'm thinking at the moment. Again, I tried Memtest earlier in the week and it did show 2 errors. Since then I've repeated Memtest multiple times (and focused on the tests it failed on) both at stock and with XMP on and I haven't been able to reproduce. I was praying it would show something more concrete but it didn't. :(

Just loaded optimised defaults with just XMP enabled and running tests again now to reconfirm.

Ran OCCT Linpack, 90% memory, no AVX and lo and behold, error detected at 11 minutes. Currently running Realbench again now.

Silent Scone@ROG
11-24-2017, 06:22 PM
XMP is an overclock, it's not going to tell you whether there are faulty DIMM. for this you need to run Memtest at complete defaults, that includes no XMP profile. This is a real fundamental, yet common misconception when it comes to memory.

Praz
11-24-2017, 06:26 PM
Hello

As noted above with default settings enable XMP and set Tweak Mode to 1. If there are still errors test each module individually using the same memory slot each time.

mikex999
11-25-2017, 03:22 PM
this xmp2 is really a fraud

PilotGW
11-25-2017, 09:47 PM
Hi everyone, thank you for your suggestions, however I think I may have fixed the issue (at least partially, more on that in a bit)...and it may have been my fault.

Last night I reseated my RAM...I've got a feeling it wasn't properly clicked in at both ends.

I put everything back to stock with just XMP. I then proceeded to run RealBench and it went for an hour and a half before I stopped it myself which is further than it'd ever gone before with XMP enabled...this was a good sign but I needed to go to bed and I needed to get Memtest going.

Loaded up Memtest overnight and it ran for 10 passes with no errors with bone stock settings.

Early this morning I restored my overclock with XMP enabled and loaded Prime95 Blend. At first it found an error in seconds, but I'd forgotten to tone down the ludicrous VCCIO voltage to 1.15v. I also raised the VCCSA voltage to 1.2v. Blend then proceeded to run for 13 hours while I was at work with no errors.

Loaded up RealBench again tonight and it passed for the full 2 hours I set it to.

I don't know whether these are flukes, but I haven't managed to pass anything with XMP on before.

Now we get to the 'partially.' OCCT Linpack still finds errors...although it did manage to go a little bit further this time. To be honest, I think I'm just going to stop tinkering now. It's been a long, frustrating week and I'm a bit tired of it now. I just want to play some games!

Thanks again for your help guys!

PilotGW
12-13-2017, 01:11 PM
Well, seems like it was a fluke after all.

I played Wolfenstein 2 and started getting 'crash dump errors.' I didn't think much of it as lots of people are getting the same error. After that I started playing Assassin's Creed Origins and it CTDs randomly. Sometimes it crashes within 20 minutes and once I played 20 hours over several days without it crashing. These are not hard lock-ups or BSODs, the game just randomly closes and Uplay asks me to send a crash report. Again, I don't know whether it's just the game. Regardless, this prompted me to go back to stress testing just to check.

So I loaded up 12 instances of HCI Memtest with 98% of the total memory allocated. It went past 1000% before I stopped it. No errors.
I then loaded Prime95 and ran a blend test for 24 hours. No errors.

At this point I was pretty confident so I loaded RealBench Stress Test as this is the one that gives me the most trouble...instability detected (7z) after 40 minutes. :mad:

Once again:

Getting rid of XMP helped it pass 2 hours
Putting the CPU back to stock, XMP enabled, with a lowered IO voltage (ridiculous at stock), instability detected after about an hour.

One other thing I tried last night out of desperation was manually entering the clock, timings and voltages from a comparable (Corsair Vengeance RED LED) but lower 2666MHz kit and that seemed to work for 2 hours.

So at this stage I think I've accepted I have one of the world's worst 8700Ks. Since I have no grounds for returning that, I'm thinking of returning the RAM and getting a 2666MHz kit instead, or alternatively, just running this set at 2666MHz.

If I just wanted to run it at 2666MHz what is the best way to go about it? Do I enable XMP and just lower the clock speed and not touch the timings and voltages, or do I use similar voltages and timings from a similar kit and go from there. Or is it best to just get a dedicated 2666MHz set?

PilotGW
12-13-2017, 11:32 PM
Okay, quick update, I've just spent the evening testing with RealBench again. This time I tried everything stock/auto (including IO voltage this time) with XMP enabled. Instability detected.

I also tried manually entering the timings and voltage with Tweak Mode 1 as suggested in an earlier post. Everything else at stock. Same thing.

I then pulled out one stick with XMP back on. For some reason I was unable to test with "up to 8GB" as it said I didn't have enough memory. I don't have this problem when selecting "up to 16GB" with both sticks in...regardless I dropped it down to 4GB and it went for over an hour before I stopped it. I then swapped the stick over using the same slot and repeated the test. This one failed within 7 minutes. I tried again and it failed after 6 minutes. I restarted the PC and it failed in 3 minutes.

To prove it wasn't a fluke I swapped the other stick back in and once again it went for over an hour. I then swapped them one last time and it failed in 15 minutes.

So, it looks like one stick is weaker than the other. The thing I can't understand is why every memory test I've done has come up clean (HCI Memtest and Memtest86).

I've just put in a replacement with Amazon. Hopefully that'll fix it. :confused:

Silent Scone@ROG
12-14-2017, 10:28 AM
I think there's a mixture of a few things happening here - for instance, you're lowering the auto values for VCCIO when enabling XMP. These values scale because the CPU can need the voltage in tow with memory overclocking. I'd recommend investing in some GSKILL TridentZ memory and returning the Corsair kit to the retailer.

PilotGW
12-16-2017, 03:45 PM
I think there's a mixture of a few things happening here - for instance, you're lowering the auto values for VCCIO when enabling XMP. These values scale because the CPU can need the voltage in tow with memory overclocking. I'd recommend investing in some GSKILL TridentZ memory and returning the Corsair kit to the retailer.

I was seriously considering getting GSKILL, they were just a little out of my price range and I already had the replacement kit on order. :(

Early indications suggest the new kit has fixed the situation. I put my overclock and XMP on right off the bat. I didn't touch VCCIO or VCCSA. VCCIO still seems a little high. If I type in the same value manually it shows up purple instead of white or yellow. I'm assuming colour means (un)acceptable ranges?

Anyway, it passed RealBench for 2 hours multiple times, which would almost invariably fail at 20-40 minutes before. I know I should run it for longer I just need to find the time to sit and monitor it. I don't like leaving RealBench unattended as it seems to do weird things with my Corsair Link. It very often kicks my fans up to 100% and once it shut my AIO pump down. Not sure if that was related to the bad memory or just Corsair Link being buggy. I'm thinking the latter.

For good measure it also passed 1200% HCI Memtest, 6 passes of MemTest86 and Windows Memory Diagnostic. I've also reinstalled Windows just in case it got corrupted somehow.

Hopefully I won't be back in another week with more problems!

Silent Scone@ROG
12-16-2017, 04:02 PM
Nice, that much coverage with HCI is fairly solid, so sounds like you're good to go :)

PilotGW
12-16-2017, 09:44 PM
Urgh...seems like I can't catch a break. :(

Was aiming to run RealBench for 4 hours and had a BSOD after 2 hours with DRIVER_OVERRAN_STACK_BUFFER relating to ntoskrnl.exe.

I've got a feeling it may be related to the Corsair Link problem in my previous post because the AIO pump started making a funny noise as if were running really slow. When it restarted, my H110i GT was not detected. Had to uninstall the 'unknown device' in device manager and let it reinstall itself.

Could it be anything else?

P-Dub
12-17-2017, 03:51 AM
Did you happen to have the CMU16GX4M2C3000C15R Corsair kit? We may have the same CPU/Mobo/Ram combo. I'm wondering because I have had random reboots after a cold boot at XMP and instability when trying to overclock slightly.

Silent Scone@ROG
12-17-2017, 08:38 AM
Urgh...seems like I can't catch a break. :(

Was aiming to run RealBench for 4 hours and had a BSOD after 2 hours with DRIVER_OVERRAN_STACK_BUFFER relating to ntoskrnl.exe.

I've got a feeling it may be related to the Corsair Link problem in my previous post because the AIO pump started making a funny noise as if were running really slow. When it restarted, my H110i GT was not detected. Had to uninstall the 'unknown device' in device manager and let it reinstall itself.

Could it be anything else?


You can always try the AVX offset function . If passing for 4 hours, setting a value of 1 should do it. Sounds more like CPU instability to me.

PilotGW
12-17-2017, 09:44 AM
Did you happen to have the CMU16GX4M2C3000C15R Corsair kit? We may have the same CPU/Mobo/Ram combo. I'm wondering because I have had random reboots after a cold boot at XMP and instability when trying to overclock slightly.

Yes, same kit. I did have occasional boot loops when I first started but after reseating the RAM it seemed to subside. However the instability remained. The replacement kit hasn't exhibited this behaviour yet and it can be pushed further. The whole thing seems a bit flakey as even the seemingly 'bad' kit passed barrage after barrage of memory tests.


You can always try the AVX offset function . If passing for 4 hours, setting a value of 1 should do it. Sounds more like CPU instability to me.

I did have a -3 AVX offset applied when I first started overclocking but turned it off as it seemed to be triggering while playing games too. I'll give it another go. 100MHz isn't going to hurt much in the grand scheme of things.

Deepcuts
12-18-2017, 05:05 PM
My 4x8 Corsair kit only works @ 3200 Mhz in Mode 2
Have you tried Mode 2?

el terrible
12-20-2017, 12:02 PM
My G.Skill tridentz 3600 mhz cl 15-15-15-35 does not work at all on XMP, I mean my computer boots just fine, I can run all kinds of tests with no errors, but when I try to actually game my computer will crash to desktop every 5-15 minutes, sometimes why whole computer will lock up and I even got a bluescreen once (haven't gotten that in years).

Currently running them at 3200mhz cl 17-18-18-36 at 1.344v and it has been stable for a few weeks with no problems.

Hoping it will get fixed in a bios update in the future

Silent Scone@ROG
12-20-2017, 03:05 PM
My G.Skill tridentz 3600 mhz cl 15-15-15-35 does not work at all on XMP, I mean my computer boots just fine, I can run all kinds of tests with no errors, but when I try to actually game my computer will crash to desktop every 5-15 minutes, sometimes why whole computer will lock up and I even got a bluescreen once (haven't gotten that in years).

Currently running them at 3200mhz cl 17-18-18-36 at 1.344v and it has been stable for a few weeks with no problems.

Hoping it will get fixed in a bios update in the future


"All kinds of test" being what tests exactly? A BIOS update is not assured to resolve anything. You can try adjusting the VCCIO and VCCSA voltages starting from 1.15v. Memory overclocking on this platform is fairly straight forward, so aslong as the user is able to POST at XMP very little tuning should be needed unless shooting for the stars (4000Mhz+)

JustinThyme
12-20-2017, 05:38 PM
Just a quick run down on dram and XMP.

Firstly....
When asking for advice/help always list your full specs. Its easiest if you go to your profile and populate the info there so you only have to enter it once. When talking about dram please list the actual part number.

XMP is an overclock. Its how the kit was binned and tested on the test platform at the factory, Its not a guarantee. The guaranteed settings is the jedec standard for your MOBO. If it says 2666, which is the Z370 strandard, that's your guaranteed speed. What runs on one board may not run on another. Some kits you pop in, click on XMP and BAM it works no worrys. Some it never works but makes for a good starting point to tweak on until it works. The OC on anything else, particularly the CPU and definitely the BCLCK will affect whether or not XMP works. Jack your BCLCK and its pretty much a guarantee you will have a headache trying to get the dram to work with it if at all.

Not all kits are alike. Bargain dram is bargain for a reason. They can write whatever they want on there for XMP values as its not the JEDEC. Think of XMP as an overclock preset.

Your MOBO and CPU also play a large part in dram stability. Dont forget the memory controller resides with the CPU.

DO NOT MIX KITS!!!I Buy what you want now. Dont buy 32GB now with the premise that you will buy 32GB more later to upgrade to 64GB. Mixing kits, even the exact same model number is a risky venture. You may get lucky and it works flawlessly but the odds are not forever in your favor on this. This forum and many others are full of dram issues that are the result of mixing kits. The higher the speed the more important this is. Sticks are first binned for like properties then tested pulled and grouped as a kit and tested again as a kit. The more sticks there are in a kit the more likely it is for things to not work as planned and the more effort it takes on the part of the manufacturer to bin the kits. Some are under the impression the sticks just roll off an assembly line and get packed up in quantities of 1, 2, 4 or 8 etc. Its not so. It takes more effort than that.

Finally and possibly most importantly.......PURCHASE RAM THAT IS ON YOUR MOBO QVL LIST!! This was you know that a particular kit was tested and confirmed to work with a particular platform.

el terrible
12-20-2017, 06:20 PM
"All kinds of test" being what tests exactly? A BIOS update is not assured to resolve anything. You can try adjusting the VCCIO and VCCSA voltages starting from 1.15v. Memory overclocking on this platform is fairly straight forward, so aslong as the user is able to POST at XMP very little tuning should be needed unless shooting for the stars (4000Mhz+)


Aida64, Windows memory diagnostic tool, prime95, Asus RealBench and cinebench all runs fine with no errors, but pretty much every single game I play at the moment crashes to desktop in minutes (Assassins Creed Origins, CS GO, PUBG, Shadow of War etc)

I don't really know anything about ram voltages because I have always used XMP on my previous builds and never had to touch anything.
But when I tried to set everything to auto and run at 3200mhz it set the timings to 17-18-18-36 and wanted to push 1.5v into the ram and I thought that was a bit much since it is only built for 1.35 running on XMP (which is 3600mhz cl 15 15 15 35) so I changed it to 1.35v and everything just works now, haven't had a single crash. I just wanted to enjoy my new rig after like 2 weeks of not understanding why all my games crashed on my new system so haven't bothered testing anymore, wanted to wait for a new bios before making any changes. Thanks for you reply though :-)

Silent Scone@ROG
12-20-2017, 07:01 PM
Aida64, Windows memory diagnostic tool, prime95, Asus RealBench and cinebench all runs fine with no errors, but pretty much every single game I play at the moment crashes to desktop in minutes (Assassins Creed Origins, CS GO, PUBG, Shadow of War etc)

I don't really know anything about ram voltages because I have always used XMP on my previous builds and never had to touch anything.
But when I tried to set everything to auto and run at 3200mhz it set the timings to 17-18-18-36 and wanted to push 1.5v into the ram and I thought that was a bit much since it is only built for 1.35 running on XMP (which is 3600mhz cl 15 15 15 35) so I changed it to 1.35v and everything just works now, haven't had a single crash. I just wanted to enjoy my new rig after like 2 weeks of not understanding why all my games crashed on my new system so haven't bothered testing anymore, wanted to wait for a new bios before making any changes. Thanks for you reply though :-)



None of those are memory stress tests. If you want to revisit it anytime, use HCI Memtest Pro, open one instance per thread and assign 90% of available memory and let the test run for at least 400% coverage.