8700k overclock issues?

IndyColtsFan

Lifer
Sep 22, 2007
33,233
446
126
I’m looking for 8700k overclocking suggestions for the rig in my sig. Currently, I have it overclocked to 5 GHz on all cores with a fixed voltage of 1.375 V and LLC at level 2. The system has been rock solid with very cool temps, never going above 60 or so in heavy gaming and only 1 or 2 cores even exceed 70 on Prime. However, I recently started playing one game (Battlefront II) that causes the system to hard lock. If I reduce the OC, it seems to play fine. Any suggestions on what to tweak? I’ve upped the voltage and it didn’t seem to stabilize it in this game. So far, only Battlefront crashes consistently in the 7 months I’ve had the system.
 

IndyColtsFan

Lifer
Sep 22, 2007
33,233
446
126
What is the AVX offset set to?
-2. This chip was rated by Silicon Lottery to be AVX stable at 4.9 GHz - they rated the chip stable at 5.1 GHz with AVX offset at -2 at 1.41 V, but I dialed it back a little. I’m not sure I want to run the chip at 1.41V 24/7.

Also, I had coretemp running when the game crashed. Temps were fine (low to mid 50s) and voltage was 1.36-1.375.
 
Last edited:

BonzaiDuck

Lifer
Jun 30, 2004
13,897
661
126
Off the top of my head, and unfamiliar with the Z370 chipset (mine is Z170), I'd first be interested if you had a BSOD with an identifiable stop-code.

That being said, find your VCCIO or IMC spec "safe-maximum" and then bump up your voltage there just a tad to see if things change for the better. You would do this with the XMP settings you have. With my Skylake-knowledge limitation, overclocking the processor or adding additional memory may require a slightly higher IMC voltage.

Also it could be useful before overclocking to make a good guess as to the stock LLC setting under "Auto." For me, I was able to be fairly confident that my Z170 board with LLC =
"Auto" actually made LLC = 5 for stock settings. In any case, you might want to adjust LLC so that the vDROOP appears to be maybe 20mV.

But with no Coffee Lake and Z370 first-hand experience of my own, I'd hope someone else has better insight than mine.

Also, the motherboard has its own limitations. The AsRock Taichi Z370, in a Tom's Hardware Benchtest and Review, shows a con of "mediocre over-clocking." The review's "VERDICT:" "Unfortunately, poor overclocking for this price class paired with poor efficiency and excess heat will make the board unattractive to many enthusiasts. "

I see you purchased the chip from Silicon Lottery, and maybe had it "re-lidded." What board did they use to test the processor? I am also a Silicon Lottery fan, and do not think the extra expense excessive if you get what you want.
 

IndyColtsFan

Lifer
Sep 22, 2007
33,233
446
126
Tom’s review was on a pre-release BIOS and was later changed with an addendum, saying it would’ve won an award had they had a fixed BIOS (page down):

https://www.tomshardware.com/reviews/asrock-z370-taichi-intel-coffee-lake-atx-motherboard,5279-4.html

The Taichi was on Silicon Lottery’s recommended list at the time, which is another reason I went that route. It is just a weird issue. I’ll have to play around with voltages some more to see if I can find the right combination.

EDIT: Believe it or not, I just notched LLC up to LLC1 from LLC2 and the game isn't crashing. Interesting.
 
Last edited:

elpokor

Junior Member
May 22, 2017
23
9
51
Battlefield 1 runs on the same engine as Battlefront 2 (Frostbite 3 iirc), and it's known for crashing overclocks that are stable elsewhere. I think it uses some sort of AVX instructions but it somehow doesn't pull all the voltage the CPU needs to do so. I've experienced it first-hand with a 6600k and 8700k, so flaky overclocks are prone to fail.

But first, the sanity checks: 70ºC in Prime95 for 1,375Vcore seems too low. I'm guessing you're using the 26.6 version without AVX, you should stress-test with AVX instructions as some games make use of them when available. Based on my own experience, if you pass the standard 10 loops of Intel Burn Test 2.54 in its "High" preset (2048MB of RAM) or "Very High" preset (4096MB), that's the kind of stability you guarantee going through hours of Prime95 (with AVX instructions). You may need to give some more milivolts to Vcore or play with LLC settings. Mind you VID=!Vcore, check CPU-Z or the Vcore reading from hwinfo64 to properly read the voltage that's coming into the CPU pins.

Second sanity check: I own the ASRock Z370 Extreme4 which should be equivalent to the Taichi in terms of overclocking as both use same layout and VRM; check https://www.hardwareluxx.de/community/f12/lga-1151-mainboard-vrm-liste-1175784.html#z370). The board sets VccSA and VccIO too damn high in Auto settings, and overshooting said voltages hurts stability. Not only it does that, but some brick-head at ASRock felt like when you overclock memory, you need to pull up aswell the VccST (absolutely WRONG: manually FIX VccST to the default voltage if you're not using LN2, plain simple).

That resulted in some funny voltage readings when I first loaded the XMP of my sticks (4000Mhz CAS19):
-VccSA went from 1.05 to 1.35V ! (Auto)
-VccIO went from 0.95 to 1.30V ! (Auto)
-VccST went from 1.0V to 1.20V ! (Auto)

I know from Skylake that 1.35V in the System Agent is not suitable for 24/7 use, but I tried to run Battlefield just for fun. CPU was running pretty toasty nearing the 70ºC region and it crashed after a few minutes. So after some days of trial and error, I ended up with a rock-solid configuration at 4000Mhz CAS16 1,43Vdram and the following CPU voltages:

-VccSA
pulled down from 1.35V(Auto) to 1.20V (in BIOS), reading 1.216V as average real voltage in hwinfo64.
-VccIO pulled down from 1.30V(Auto) to 1.15V (in BIOS), reading 1.165V as average real voltage in hwinfo64.
-VccST pulled down from 1.20V(Auto) to 1.00V (in BIOS), reading 1.008V as average real voltage in hwinfo64.

Now the 8700k is cooler due to the reduced voltages, and it's stable through everything I can throw at it. Here's the memtest with the final timings I manually nailed in the BIOS:


So for your 3200Mhz RAM kit at stock XMP I'd say you could potentially even lower VccSA and VccIO more than I did (1.15V and 1.10V is my guess), and if you have those voltages at Auto they're probably too high and/or causing instability. VccST should always be fixed to 1.000V unless you are under liquid nitrogen cooling.

I hope that helps!
 
  • Like
Reactions: Headfoot and Campy

IndyColtsFan

Lifer
Sep 22, 2007
33,233
446
126
Thanks for the info! What I may not have mentioned is that my CPU is delidded.

On Prime, what I mentioned was that only 1 or 2 cores exceed 70 - I believe one hits 75. It was using AVX for sure because the clock did drop to 4.8 GHz during the run. Now, I have not rerun Prime since notching up LLC; I would expect temps to go up a few degrees to the point where all cores would cross 70 and maybe a couple would flirt with 80.

At this stage, I’m going to stop tinkering unless I see more crashes. I’ll sacrifice a few degrees in temp to get stability.
 

Campy

Senior member
Jun 25, 2010
664
116
106
Fantastic post by elpokor!

I would also recommend to check SA and IO voltages and make sure they are not too high on auto, as this in the worst case can degrade your IMC.
 
  • Like
Reactions: elpokor

IndyColtsFan

Lifer
Sep 22, 2007
33,233
446
126
Fantastic post by elpokor!

I would also recommend to check SA and IO voltages and make sure they are not too high on auto, as this in the worst case can degrade your IMC.
IO was set to 1.2 IIRC. I don’t recall the SA setting.
 

ZGR

Golden Member
Oct 26, 2012
1,829
250
126
Does Silicon Lottery test their chips with high speed memory?
 

elpokor

Junior Member
May 22, 2017
23
9
51
Does Silicon Lottery test their chips with high speed memory?
according to their QVL yes, up to 3200Mhz https://siliconlottery.com/pages/intel-coffee-lake. It shouldn't be a problem with Coffee Lake unless you're aiming over 4200Mhz though, I've tested two 8700k and Intel's claims of "better memory overclocking" seem accurate. Both did 4000Mhz with tight timings at the same voltages: 1.15VccIO and 1.20VccSA for one SKU and 1.14VccIO/1.20VccSA for the second. I haven't owned a Kaby Lake CPU, but in Skylake you needed that kind of voltages to get 3200Mhz stable

IO was set to 1.2 IIRC. I don’t recall the SA setting.
take a look at this topic http://forum.asrock.com/forum_posts.asp?TID=6860&PN=1&title=z370-extreme4-xmp-bug . It's a known issue that ASRock overshoots voltages when you turn on XMP, and that has caused some problems with the profiles. You'll notice there's 10 pages of people reporting that turning down the voltages does indeed work. I know for a fact that overvolting IMC or System Agent does cause instability, been there done that... :p Trust me (and 10 pages in ASRock forums): you need to manually fix1.10VccIO/1.15VccSA/1.00VccST to be stable at the speed you're aiming for.

But given that you own 3200MhzC14 sticks, I'd suggest you to copy-paste every single subtiming and voltage from my screenshot above: it's a free cheat-sheet to get a >90% efficiency in memory overclocking with your same setup (ASRock Z370 same physical board and same Samsung B-die ICs). Max theoretical throughput at 4000Mhz memory is:
4Ghz x 64-bit bus x 2 (double data rate) / 8 (bits to bytes conversion) = 64GB/s

now here's the performance of the slightly-tweaked XMP with Auto voltages:


and this is after manually tweaking the latencies and voltages to the values shown in my previous post:


So it's absolutely worth the time spent tinkering, the Auto settings are not good enough on these boards we're dealing with. Let me save you some time and hassle and give a try to any of my suggested settings :p
 
  • Like
Reactions: ZGR and Headfoot

ASK THE COMMUNITY