x299 not stable using manual voltage

jyi786

[H]ardness Supreme
Joined
Jun 13, 2002
Messages
5,511
Just as a preface, I'm a quite experienced OC'er...but a relative n00b to Skylake-X OCing. I know that the x299 platform and Skylake X is a beast to try to OC. Between the extreme multitude of voltages, multipliers, AVX, etc., it can be quite daunting in trying to attain a stable OC.

Currently, I've OC'd my 9980XE to 4.5Ghz @ 1.163V (manual), with AVX offsets at 12 and 12. I ran through a battery of stability tests, including Realbench, AIDA64, Cinebench, and Furmark. I was using it this way for a month. There were no signs of instability. Or so I thought...

I decided yesterday to try OCCT, and the system shuts off immediately. Same for Prime95. After 2 entire nights of trying to figure it out, I said **** it and finally set all the CPU and its related voltages (Vcore, CPU input, cache, etc. etc.) back to Auto.

Lo and behold, the system was completely STABLE. Anytime I do anything with manual voltages, doing any type of OCCT or Prime 95 testing immediately causes the system to shut down, which is OCP (over-current protection). I found the switch in the BIOS to prevent it from happening. This switch is ignored if SVID control is enabled.

Through all my testing, no matter what I do, Auto voltage is stable, whereas anything I ever do with Offset, Adaptive, or Manual triggers an instant OCP shutdown.

For giggles, I set everything back to defaults, made sure Auto voltages were set, and proceeded to set the AVX to 5 and AVX 512 to 5, and ensured that the switch to prevent OCP was set to 1023.875A (the default setting of Auto or a too low value is what causes the OCP shutdown). This yields 3.9Ghz for an AVX 512 operation. Doing OCCT with Small Data Set and AVX 512, I was able to run for 20 minutes straight with the system completely rock solid: no shutting down. The CPU temps and the VRM temps were both 102C. Granted, it's only 20 minutes, but I'm 100% sure I can push it and find stability, since it's running this way on AVX offset 6 and AVX offset 10 right now (I'm just being overly cautious and conservative). I then proceeded to stress test for 12 hours with OCCT and Realbench with no issues. It's rock solid at this point.

This was completely surprising to me. I usually know that you can find a stable OC using manual voltage, or with the new architectures, adaptive or offset, but any combination of adaptive or offset I use results in an almost immediate random OCP shutdown. Leaving the voltages on Auto brings back stability and prevents the system from shutting down. That goes for any type of CPU load: AVX, AVX 512, non AVX, mixed, whatever.

What could be causing this type of behavior? Here is a pic of the voltages at full AVX 512 load @ 3.9Ghz.

20191111_050707.jpg

Here are my rig specs, shamelessly lifted from my first WCing thread.

---------------------------------------------------------

https://hardforum.com/threads/my-first-watercooled-build-9980xe-rtx-2080ti-etc.1987324/

CPU: Intel 9980XE
Motherboard: Asus WS x299 Sage Pro/10G
RAM: 128GB G Skill RGB Z
GPU: MSI nVidia RTX 2080 Ti Seahawk EK
GPU support bracket: CloverTale Graphics card support bracket
Sound: Sound Blaster Z PCI-E
PSU: Seasonic Snow Silent 1050W
Case: Fractal Define R6 USB C
Triple monitors: Asus PG279Q (x2), BenQ BL3201PH (x1)

I went with 1/2" tubing for my build. The watercooling specs are as follows:

Radiator: MO-RA 420 Pro
Fans: Noctua NF-A14 PWM fans (x18) <--- [yeah that's not a typo]
Reservoirs: Heatkiller Tube 150 (x2)
Pump: Watercool D5 Vario (x2)
Tubing: Mayhems 1/2" clear tubing
Fittings: All XSPC compression fittings
Valves: XSPC black chrome (x2) [for outlet]
Liquid: Plain distilled water, no color
Biocide: Mayhems Extreme biocide

---------------------------------------------------------
 
Last edited:

Monstieur

Limp Gawd
Joined
Jun 10, 2011
Messages
171
Everyone OCs X299 wrong. The SVID of bad cores at maximum frequency on these CPUs is ~1.35 V, and your Adaptive voltage will be ignored if it is below the SVID. Thus if you overclock, you are actually doing it at stock voltage and it will be unstable. The correct way to OC is to use Adaptive + Offset mode, or plain Adaptive / Override mode with a voltage above the highest SVID.

There is a lot of misinformation regarding Adaptive + Offset mode, especially when combined with AVX offsets.
Adaptive voltage raises only the minimum voltage while Turbo is active (any frequency above base such as AVX offsets, not just the maximum). This way the CPU can idle at stock voltage (which is below the minimum configurable Adaptive voltage of 1.2 V).
Offset voltage is added at all frequencies, including idle.

The formula is:
Adaptive voltage + Offset = Minimum stable AVX-2 & AVX-512 frequency voltage
SVID + Offset = Minimum stable non-AVX frequency voltage

The SVID table is not designed for AVX loads. For example:
45x non-AVX has a stable SVID of 1.200 V.
45x AVX-512 OC has the same SVID of 1.200 V but will crash. It needs 1.225 V.
49x non-AVX OC has an SVID of 1.350 V but will crash. It needs 1.375 V.
All of these voltages are for the worst core with the highest voltage requirement.

Adaptive voltage = 1.200 V and Offset = +0.025 V will result in a stable OC.
You can also set Adaptive / Override voltage = 1.375 V which is stable but overvolts the AVX frequency heavily, as well as the good cores at the non-AVX frequency which have stable SVIDs below 1.375 V.

First stabilize the AVX frequency with just Adaptive voltage. Don’t use Offset yet as it increases the idle voltage. Then stabilize the non-AVX frequency using Offset if required. If you increase Offset, reduce Adaptive voltage accordingly.
 
Last edited:

jyi786

[H]ardness Supreme
Joined
Jun 13, 2002
Messages
5,511
Unfortunately, I've found that auto will never work. Although I've stabilized it to hell and back on auto voltage, a new problem cropped up:

Instability at idle.

This includes using any combination of Adaptive and/or Offset. This to me is worse because it's not something I can control. Even using offset is not enough to keep the voltages at idle high enough for the computer to not crash at random.

I actually settled on 1.185V for all cores, and the system is completely stable at idle and full load. The only downside is that I had to lower the AVX multipliers like insane. I believe it's AVX2 -12 and AVX512 -13.
 

Monstieur

Limp Gawd
Joined
Jun 10, 2011
Messages
171

jyi786

[H]ardness Supreme
Joined
Jun 13, 2002
Messages
5,511
The default AVX frequencies are quite low on the HCC parts - typically in the low ~3.0 GHz, so AVX offsets of -13 are normal. The auto setting for AVX offsets do not work on most motherboards and result in instability when voltages are on auto.
Yeah, I've settled on 4.4. Completely stable. Looks like I did it right. Oh, and between now and then, I've also gotten the VRMs under water. It gets no hotter than 66C at full AVX 512 load.
 
Top