Help me stabilize a 8700K / Z370 OC?

[tacticurn]

So, I have for a week tried to hit a stable 5,0 GHz OC on my 8700K with Asus Strix-E Z370. I withdrew to 4,9. I've had long, stable runs at 1.31 VCore (Manual) on this clock.
Then suddenly a freeze on Desktop in an idle moment, like opening a browser or typing something in a document after HOURS of torturous work in benchmarks and games.
I am baffled.
Temps are low. During IntelBurn cores sometimes peak at 83°C but generally ranges from 65 to 75. During gaming way less.

I gave up on adaptive and run with Manual Vcore of 1,31
I have set LLC to 5
XMP is set, to 3600-16-16-16 36. This affects various voltages of the DRAM like VCCIO and System Agent I think. The Dram gets 1,354V
CPU SVID Support is Disabled
SVID behaviour is set to Typical Scenario (but not sure how much it matters since I disabled SVID?).
C states etc. are enabled.

I disabled Asus Core Enhancements.
All core synch to 49
AVX negative offset 1.

Cpu core/cache current limit max. to 255.5
Ring down Auto
Min max cpu cache ratio Auto


So, do I add another 0,01V and see what happens at 1,32? Sounds high for a 4900 GHz OC doesnt it?

 

[ebivan]

Settle for 4.9 or give it more voltage. As long as temps are ok, more voltage should be ok. But settleing with lower voltage and lower clocks will probably be much better in therms of efficiency.

 

[nguyen]

Vcore at around 1.4V is safe on Intel 14nm parts
Vccio 1.05V
Vccsa 1.15V
CPU Cache Ratio 45
Disable all power saving feature like C-States and Speedstep should help with stability.

 

[GerKNG]

Disable everything except turbo boost. give a bit more Vcore and you#re fine.

VCCIO VCCSA (try 1.15V for both.. don't worry.. all the "do not go above 1.15v or 1.2v are pretty much pointless since people just reply with what another person said. below 1.4V for both is still "fine" but above 1.3V is not worth it and does normally not help with stability at all.)

idle crashes can be due to low or way too high VCCIO/SA.

try 1.15 for both, if it's still crashing at idle go up with the vcore.

 

[tacticurn]

Thanks for your suggestions. I have lowered VCCIO and VCCSA a bit and testing now for stability.

 

[chrcoluk]

Just a heads up i discovered on my z370 board (with 9900k), a too high vccsa was causing i/o crc errors on both m.2 and SATA ssd's. Ram was fine, so it was a form of silent corruption.

 

[Caring1]

Just a heads up i discovered on my z370 board (with 9900k), a too high vccsa was causing i/o crc errors on both m.2 and SATA ssd's. Ram was fine, so it was a form of silent corruption.

Was that throwing up any A2 errors on the Motherboard LED readout?

 

[storm-chaser]

What was going on with your adaptive vcore? Was it not up to the task? Reason I ask is because adaptive vcore is really the way to go should you choose to overclock (dynamic mode) your turbo configuration, because it allows your clock speed to "float" a little bit and draw less voltage, even in the upper turbo limits. It's also going to bring the voltage way down when not under load, again, contributing to a cooler running CPU. I would recommend this type (id go as far as to say it's superior to fixed OC) of overclock much much more than just an override / fixed ratio overclock where you are pumping a stead state 1.4v or whatever to the CPU at all times. Reason is very simple. A static overclock wont allow your CPU to boost for 1-2-3 threaded workloads, it's always going to be stuck at whatever you set it to across ALL cores. Turbo OC can also be higher performing than static if your chip can handle the upper limit in terms of clock speed. For example. my static overclocked 5.0GHZ 9600 KF netted me a score of 615 in CPUz single core results. Once I switched to turbo overclocking, I set my ratios to x53 x53 x53 x52 x51 x50. This improved my single core result to 636 due to turbo boost, and at the same time your are NOT giving anything away, because you still have the fallback to 5.0GHz when you encounter a heavy workload. In other words, you get to have your cake and eat it too. In other words best of both worlds: blisteringly fast single core performance while at the same time remaining = to a fixed ratio OC when all cores get loaded up.

Another thing to try if you are running into thermal issues you might want to reconfigure your LLC setting. Start from the bottom, which adds the least amount of voltage under load, and slowly keep bumping up your LCC until the OC is stable. If you can get your chip stable on the lowest possible LLC setting, you will see a substantial decrease in temperatures in some cases.

For example, I set my LLC two notches up from the bottom, which gives my a vcore of 1.380 when loaded up. With LLC, only use what you need and nothing more.

This adaptive vcore setting combined with LLC tuning allows me to run all my case / rad / CPU fans at 800 rpm (essentially idle) and they don't spool at all. The only thing that spools up is the water pump, and that is nearly silent anyway. If you hit it hard with a load it's totally rock solid, and despite all fans at idle, temps are still very good and well within the margins. I'm at about 170* in this configuration. So the rig is completely silent right now, I never thought I would the day!

 

[chrcoluk]

Was that throwing up any A2 errors on the Motherboard LED readout?

No idea I didnt check it.

I was getting on older sata ssd's (ones that have crappy error correction) SMART crc errors.
My 970 evo was getting "data and integrity errors".

When I lowered the voltage both issues vanished. What made me try it as I realised the problems started after I started using a higher voltage. One of the strangest problems I have had in 20 years of using PC's.

 

[elghinnarisa]

I run mine at 1.3v adaptive with a offset + 0.045v
DRAM 1.36v
VCCIO 1.2v (set by enabling XMP)
VCCSA 1.2v (set by enabling XMP)
LLC set to "6"
SVID Base-case scenario
x50 ratio limit

In a BIOS updated for my Maximus X hero they added so enabling XMP also increase VCCIO and VCCSA to 1.2v per default, since it helped with stability (or so the changelog said)
it goes up to 1.38v at full load with those settings and i personally havn't had any stability issues since i got those settings down, been running like that since a few months after I bought it, which was at release.

Also worth noting that on my motherboard the default auto voltage at stock resulted in 1.43v full load.

 

[chrcoluk]

Asrock also bump it on XMP, but many users reported on their own forums they bump it far too high, I think some of these board vendors are too concerned about people making the super elite 4000+mhz sticks stable whilst forgetting most people have slower ram and wont need such high voltages.