Or did I just lose the silicon lottery?
 
I'm trying out Overclocking for the first time, running an i5-8600k on an EVGA Z370 FTW with an EVGA CLC280 for cooling. The only way I can seem to run Prime95 small FFTs for more than a few minutes is at 4.7ghz WITH the AVX set to 2, so it's running at a real-time observed 4.5ghz. I have BLCK set to 100, and voltage set to adaptive with no offset, and vdroop turned off. For 4.7 I can keep it under 1.35, but exact voltage I've played around with a bit because the other issue is the voltage doesn't ever seem to do what the BIOS says. Even set to override, my voltage will drop below the override value, or shoot WAY above it. I've tried running 4.9ghz with override set to 1.4v but it would shoot way past 1.45v, even coming close to 1.5v at times which totally destroyed my temps.
 
I have yet to experience any instability in Windows itself, but what seems to happen each time in Prime95 is that within a few seconds of the small FFTs firing up core 4 (and sometimes core 2 shortly after) will post an unexpected return and the worker will stop. I'm also seeing some pretty high temperature overruns, I'm guessing due to the voltage spikes? Idle is usally about 31c, at 1.35v small FFTs it's usually high 80s to low 90s. But it can shoot well into the high 90s at 1.4v (or around there I guess since I can't seem to stabilize voltage).
 
Are there any other settings I can test, or should I just admit defeat stay with 4.7ghz?

Cool GTX
 Web search:
  "i5-8600k & EVGA Z370 FTW OC Guide"  (remove "")
 
   "i5-8600k & OC Guide"

   "EVGA Z370 FTW & OC Guide"


 
 
Did you see this ?

Yes, those are great guides (Edit: and specifically the ones I used to try and achieve OC)! As you can see I have tried those settings and achieved a different result. Those seemed the only two major review guides I could find for this specific board, and since the EVGA Bios settings are a bit different from others (no LLC listed ect) I can only follow guides such as so far (although I did as far as I could). For the stability testing I used guides such at that seemed pretty standard, although there is some controversy it seems over whether Prime95 AVX testing is really needed. I seemed to be able to run x264 without issue at the same settings. Perhaps that is the difference that I am getting from those two guides, that they weren't running small FFT AVX testing?


Hence why I'm asking if there is some further setting specific to the EVGA board or if I've reached the limits of my processor. Particularly, I'm wondering about my voltage fluctuations even when set to manual, which seems to be an issue specific to the board (not the processor) which made this seem like the best place to ask. Was I incorrect?

Sajin
Don't use adaptive voltage. Use a static fixed voltage. Make sure you disable vdroop if the setting is available.

Thanks! I've definitely been avoiding vdroop like the plague. I can definitely give static fixed voltage a shot again, but do you know how to get it to stay static? As I mentioned with static voltage set to 1.4v and vdroop disabled it would still shoot up to 1.45 or even higher running 8k FFTs which immediately blasted the thermals and stopped one or more workers.

Sajin
If you're using a static fixed voltage it should stay at exactly what you set it to. Very odd that it increase to 1.45 or higher when loaded.

Yeah, I'm not sure. It seems to stay pretty fixed until I throw those small FFTs at it. I've just talked to someone on a different forum running the same processor and MOBO and they ran into the same voltage jumps. It appears there are no LLC settings in the BIOS? That said, they were able to achieve 5.1ghz with only 1.287v so ultimately it would seem I simply lost the silicon lottery badly.

This is a problem I've complained several times and submitted tickets about. I don't know why EVGA ignores it.
 
Vdroop "disable" will add something around 80mV when under a normal "full" load, non-AVX. When you have an AVX load, intel spec is to add more voltage, so it stacks that on top of EVGA's vdroop compensation and I've seen over 120mV added when under AVX loads. There are no other settings to effect load-line voltage control.
 
EVGA's implementation of vdroop compensation is very dangerous because you can go from a very reasonable setting of vcore if you say, input 1.3V static, and under load see voltages way above 1.4V, which is getting to the point of degrading your chip.
 
Also, this board pumps system agent and IO voltages to borderline unsafe levels too, when selecting XMP. They should be VCCSA=1.05-1.15 and VCCIO=0.95-1.05. If you select XMP EVGA board blasts them to 1.25V, again getting close to unsafe levels and may start degrading the IMC on your CPU. Make sure you set these manually, because you CANNOT trust the EVGA board to ever use appropriate "auto" settings.
 
Source: I have a z370 FTW board.

satchmo0016
This is a problem I've complained several times and submitted tickets about. I don't know why EVGA ignores it.
 
Vdroop "disable" will add something around 80mV when under a normal "full" load, non-AVX. When you have an AVX load, intel spec is to add more voltage, so it stacks that on top of EVGA's vdroop compensation and I've seen over 120mV added when under AVX loads. There are no other settings to effect load-line voltage control.
 
EVGA's implementation of vdroop compensation is very dangerous because you can go from a very reasonable setting of vcore if you say, input 1.3V static, and under load see voltages way above 1.4V, which is getting to the point of degrading your chip.
 
Also, this board pumps system agent and IO voltages to borderline unsafe levels too, when selecting XMP. They should be VCCSA=1.05-1.15 and VCCIO=0.95-1.05. If you select XMP EVGA board blasts them to 1.25V, again getting close to unsafe levels and may start degrading the IMC on your CPU. Make sure you set these manually, because you CANNOT trust the EVGA board to ever use appropriate "auto" settings.
 
Source: I have a z370 FTW board.

 
Yeah, this is crazy! It took a lot of fiddling, but I think I found the only voltage I can work it at. Set to 1.300v manual override w/ no vdroop I can run stable at 5ghz, as long as I set the AVX offset to 2 (so really just 5.8ghz for Prime95 and x264). Anything less than that and one of my cores stops working under small FFTs, any more than that the voltages spikes also kill of one or more cores. At 1.300v override no vdroop I am seeing mid FFT testing ramp the voltage to 1.34-1.36, and 1.39 for small FFTs. Mid FFTs and better I can run low to mid 80s, but small FFTs at 1.39 I'm seeing temps around 95c. This is with an EVGA CLC 280 for CPU cooling and running all fans at max. Seems like the best I'm going to get with this board/processor combo.
 
Thanks for the tip on the system agent and IO voltages! Sure enough, they were around 1.25v with XMP on. I set them manually which still ran stable, although I noticed both were still running over what I set them to. I had to lower it a bit to keep the overshoot within acceptable range.
 
Honestly, I don't get it. I just recommended this board to a friend who doesn't want to overclock (I still stand by that), and it's totally perfect for me in all regards except when it comes to overclocking. The thing advertised for and that I bought it for. Why does override not override? Why is there no way to manage the voltages outside of testing until you find the overshoot and then setting it that far under? Too often that means setting it so low that you can't overclock. I'd understand if it was going above to try and maintain stability, but it isn't. I tried setting the voltage override high and enabling vdroop which did cause it to no longer overshoot the override, but then it just undershot it and crashed all the same. Auto seemed to be the worst of both worlds honestly. I'm not sure what the intention was here or if there is, per my title, some sort of hidden setting out there for it. I'm running EVGA MOBO, EVGA AIO for the CPU, EVGA graphics with a EVGA AIO on that, and an EVGA power supply. Basically everything but the case, so if there's any setup that should be able to OC the Coffee Lakes this was built for this should be it. I think I finally got as far as I can take this CPU, but man it isn't easy!

satchmo0016
They're most certainly not getting 5.1Ghz at actual vcore that low. The EVGA board adds a lot of voltage behind your back with vdroop - disabled, and even more on AVX loads. Check my comment above.

They're getting exactly the same issue we are, 80-120mv overshot on the voltage. I believe 1.287v is what they have their override at, but it'd be 1.37-1.4v. They said it took about 10 hours over a few days to dial that in. Just a stranger on the internet though, so who knows. But they did seem to confirm the same issues.



 

To add in my two cents, my i5-8600k is at 1.35 V override (ie fixed).
Under typical high loads (ie Cinebench), the voltage only goes up to ~1.4 V, but under AVX heavy loads, the extra spike from having to process AVX pushes voltage up to ~1.45 V.
TMPIN1 is the VRM temperature I believe. My CPU is delidded with Thermal Grizzly Conductonaut applied, cooled with a Corsair H100i v2.
While 1.45 V is a bit high, it's not exceedingly high since Silicon Lottery sells them with voltage settings such as 1.435 V for the 8086k.

If you are interested in trying the Intel Linpack benchmark, you can acquire it at 
Extract the zip file, open a Powershell or Command Prompt and go to that folder location, and run the file.
I usually do 10000, 20016, and 40016, which take about 7, 30, and 120 seconds each, respectively.