Massive vCore dips when running AVX on Z370 + 8086K?

Ended up winning a free 8086k on the giveaway a while back. So I finally got around to a delid/relid and installed the processor yesterday. I'm having and issue where the vCore will dip a large amount when running AVX processes so I'm having a hard time running stress tests to verify a solid OC.

First, specs:
EVGA Z370 Classified K mobo
32GB G.Skil 3600MHz DDR4 (4x8GB)
Seasonic Focus Plus Gold 850w
EVGA 1080Ti SC2 Black
XSPC Raystorm Neo CPU block
EVGA water block on 1080Ti
XSPC D5 photon v2 pump with 270ml resovoir
XSPC RX360 360mm radiator running XSPC EC6 coolant
6x Noctua NF-A12x25 fans on radiator in push/pull

i7 8086k settings:
CPU Multi = 50
Ring = 50 (I've tried from 40 to 50 trying to see if this changes anything)
vCore = Adaptive at 1.20v
VCCSA = 1.26v
VCCIO = 1.26v
vdroop = auto/enable

RAM is set to XMP 1 = 3600MHz at 1.35v

Temperatures in AVX loads (for as long as they will run) are around 65-70c with an ambient of 83f, and run around 45-50c in games/other processes at the same 83f ambient.

Now, when I boot into Windows all is well and stable. Sitting at desktop and opening windows/programs the CPU will clock at 5GHz and run anywhere from 1.25v to 1.31v, same with non-AVX processes, games, etc.

When I launch up Cinebench, prime95, or any other process that can/does use AVX instructions the vCore will drop to 1.108v - 1.15v and will fail after a few minutes. I've tried changing the adaptive voltage all the way up to 1.45v and it will still drop to 1.15-1.2v when running an AVX threaded process. I've also tried changing the AVX offset from 1 to 3 with the same results, although the tests don't fail when AVX offset is set.

The same issue happens when I set the vCore to static/override. When turning off the vdroop, I have to run vCore at 1.2v with a -180mv offset for the voltage not to spike up to over 1.5, but the computer is not stable in normal processes with the voltage dropping so low at lower loads.

The EVGA BIOS does not have a setting for LLC, so that's not something I can mess with. To get to the main point; is there anything else that can be done to stop the vCore drops when running AVX loads to properly test stability? I have no stability issues in games or normal processes at 5GHz with my current settings, but I am unable to run an AVX stress test successfully due to the vCore dropping so much during AVX stress. Should I just not worry about AVX stress tesing and run an AVX offset of 1 or 2? (AVX offset 1, which runs AVX loads at 4.9GHz, doesn't seem to fail even at 1.1v)

Thanks for any advice/assistance and taking the time to read this long post.

AVX offset is just a crutch imho. For instance Linux will trigger that offset 100% of the time. So will some Windows games. It seems that true 5GHz stability is a lot harder than many posts would lead you to believe. Just look at the voltage that SiliconLottery suggests, 1.400v for avx offset -2. I wound up at 1.37v LLC5 for 5.0 stability with AVX but then it throttles a bit in Prime95 even with delid/relid, but I don't care because I don't use avx/fp intensive programs like Prime95 for hours on end. If I was willing to let avx crash then the voltage could go down a lot.

Well, I can't say what the issue was.

I can say this now. I was messing around with the Core PLL voltage and got to a point where the computer wouldn't boot at all. I kept getting a C error on boot on the motherboard which indicated a "motherboard failure". After spending a few hours, several swapped processors and power supplies, it started working again. Now the voltage is doing the exact opposite and I had to severely change my settings. After all this I'm now using the following:

Multiplier = 50 - all cores
Ring = 50
AVX offset = 0

vCore = Adaptive @ 1.2v with a -120mV offset (was using 0 before) | With this setting I'm seeing 1.215-1.23 under normal loads and 1.3-1.315 under AVX loads
VCCSA = 1.26
VCCIO = 1.25

Same vdroop and RAM settings.

I'm not sure if there was some issue with the board and swapping out parts and power supplies "knocked" it loose from whatever the issue was, but now it seems to be working as it "should" in terms of how the voltages change under different loads.

Will be testing the following for stability after whatever sorted out the voltage ordeal.

So far it is stable after 15 minutes at 85F ambient but will let it run overnight for testing

Current settings:
Multi:
1 Core = 52
All cores = 50
Ring = 50

vCore = Adaptive @ 1.2v with -105mV offset | Under AVX this works out to 1.32-1.328v and at 85F ambient has an average core temp of 86c with core 3 hitting a high spot of 93c and all other cores under 84c | Under non AVX load it only hits around 55-65c
VCCSA = 1.26 (actual 1.256)
VCCIO = 1.25 (actual 1.256)

I've always been horribly unlucky with the silicon lottery with every chip I've had, so I'm expecting I'll end up having to settle for either higher voltage or a bit of AVX offset if I'm going to be using AVX in any way.

When I reseated my delid, I got CC error, that's memory. Somehow all the pins weren't making good contact, would only boot with one stick until I reseated the cpu again.

With no LLC you need to set your vcore high enough to cope under load.
Not too high that it causes a high excess of wear over time.
There is no problem running a higher vcore if its within safe limits, it will achieve the same end goal as LLC but use a little more power at idle+lower load.

If adaptive proves troublesome try a permanent voltage.
Perhaps try that first so you know what you are aiming for when setting adaptive.

ps
vccio and vccsa should be kept as low as possible while keeping your ram stable.
These add to the CPUs heat and too high a setting can cause damage to the memory controller.
Bear in mind that what you set may drive a much higher voltage, my motherboard does this on both of them.
I have to set them lower to get values that wont cause damage.
(This is while overclocking DDR4 3000 to 3766MHz)

If you are not overclocking your ram, use recommended settings for vccio and vccsa or leave them at default.

xorbe said:

AVX offset is just a crutch imho. For instance Linux will trigger that offset 100% of the time. So will some Windows games. It seems that true 5GHz stability is a lot harder than many posts would lead you to believe. Just look at the voltage that SiliconLottery suggests, 1.400v for avx offset -2. I wound up at 1.37v LLC5 for 5.0 stability with AVX but then it throttles a bit in Prime95 even with delid/relid, but I don't care because I don't use avx/fp intensive programs like Prime95 for hours on end. If I was willing to let avx crash then the voltage could go down a lot.

Click to expand...

Exactly my experience with 9900K. 5ghz 1.160v tested 1 hour in real bench and 2 hours bf1. Can't even boot linux until 1.192v and compile gcc in 10 loops until 1.232v.

Where can I get more information on why Linux always triggers AVX?

mouacyk said:

Exactly my experience with 9900K. 5ghz 1.160v tested 1 hour in real bench and 2 hours bf1. Can't even boot linux until 1.192v and compile gcc in 10 loops until 1.232v.

Where can I get more information on why Linux always triggers AVX?

Click to expand...

I am guessing in the context switch, but that's just a wild guess.