5930K ASUS X99-A OC Results

[hemirunner426]

It's been a very long time since I've posted here.

I'm here to post some moderate OC results. I'm hoping these settings will help some others start with a baseline and satisfy a few questions along the way.

OC Results:
BCLK: 100MHz
Cache: 4100MHz
Sync All Cores: ON
Multipler: 44x
Memory Freq: 2666MHz @ 1.20v
vCore: Adaptive Turbo: 1.320v Offset: Auto
vCache: Offset .25, max 1.20v AUTO voltage
VCCSA: Auto @ 1.16v
VCCIN: 2.00v
CPU SVID Support: Disabled
CPU Spread Spectrum: Disabled
PLL overvoltage: Enabled
Attempt Fast Cold Boot: Enabled
Attempt Fast Boot: Enabled
LLC: Auto (Will add approx. .08v to vCore at high loads to a total of 1.328v)
VRM Spread Spectrum: Disabled
Active Frequency Mode: Disabled
EZ XMP/EPU/TPU switches all off

Everything else is left Auto/Default

Temps:
CPU: Prime95 27.9/HandBrake/Aida64 will max around 75C on a custom loop and approx. 80F ambient temps. Aida FPU only test will spike it to 80-85C due to the AVX2 instructions.
MB VRM: 80C at Prime95 loads

I didn't win the silicon lottery this time. It took a lot of work to crest the 4.4 point on this chip. Here are my questions:

1. CPU Input voltage... the above is what it took to get stable at 4.4GHz at 1.32v. I'm aiming for a 24/7 O/C. Is that a reasonable voltage?
2. VRM temps at full load look a little high, but not too crazy. Due to the passive cooling on the VRM heatsink and the W/C I've no doubt lost some airflow over the area. Removing the case side doesn't drop the temps at all so it looks like I have good case flow. I'm tempted to say those VRM temps are OK. Typical loads bring about 65C. What are your thoughts?
3. VCCSA @ 1.16v is common at 2666MHz? This is the auto voltage setting. I may mess with this as I start to push the memory further.
3. What are your thoughts overall?

4.4 will be the max I'll push it. The water cooling setup I built wont effectively handle the higher heat output during these hot summer months (Phoenix, AZ) without some sort of chiller. I didn't exactly plan for 300W of heat dissipation.

 

[Woomack]

1. All depends from chip. 1.32V is not low but still reasonable for 24/7. Let's say you can try 1.35V @4.5GHz but if you had to set 1.32V for 4.4GHz then it probably requires something near 1.37V for 4.5GHz.
2. VRM should be fine. 65*C looks high but usually are no bigger problems up to 100*C.
3. In most cases you can keep VCCSA at stock up to 3000 memory clock. Stock is something near 0.9-0.95V. If you set it to 1.0V then will be just fine. Motherboards are often setting higher voltage than required due to compatibility with all possible configurations. 1.1V+ can be required for 8x8GB at higher clock but I don't think any 4 module memory kit needs more than 1.0V. Most my 4x4GB memory kits were running fine up to 3200 on 0.95-1.0V VCCSA.
4. My thought are only that if you are satisfied with overall performance then all is fine. My brother has [email protected] and he doesn't need anything more to use every application he wants even though he has quite good CPU and could OC it much higher.

These 6 cores don't really need much better cooling than 4 cores from Haswell/Skylake series. Safe temps are up to ~90*C under load and maybe higher but I've noticed that some CPUs are generating errors above that point and some tests are failing.
I don't know if you have option in BIOS to raise maximum operating temp. On Rampage V I'm setting it to 95*C while standard is 70 or something near. Anyway at your settings it shouldn't throttle even when temps will go higher. 80*C+ in AVX tests is still good and should be fine for 24/7 work.

 

[EarthDog]

You dont need to touch input voltage... leave that on auto or set it manually to its stock value. You also do not need to touch PLL Override voltage.

Otherwise, +1 to what Woomack said.

 

[hemirunner426]

Input voltage was the only way I could get it stable at 4.4 around 1.3v. 1.92 was the auto default. I stress tested at 1.92, 1.95, 1.98 and didn't find stability at 1.32 until 2.0v.
VCCIN seemed to be the key.

I'm not sure if that is the norm or if that is on the edge of being 'too much'.

@Woomack:
The Asus X99-A gives you an option to override tjmax, which defaults at 100C. It works too, as I had my CPU throttle when I had a pump failure on an H240X.

 

[EarthDog]

Its fine, fine, but, doesn't make much sense.

 

[EarthDog]

Its fine, fine, but, doesn't make much sense.

 

[hemirunner426]

Its fine, fine, but, doesn't make much sense.

I would agree too. It is strange. I suppose that's why I'm asking for some hypothesis. It certainly is interesting, unless I'm missing something in my settings that would allow me to reduce VCCIN. I'm not entirely sure!

I've read other people saying they've increased input voltage to reduce vCore slightly. Maybe that is the same effect here just a bit on the extreme? For instance, perhaps I can reduce input voltage to somewhere around 1.92 and add vCore, say 1.37v to stay stable at 4.4GHz. I never did break 1.35v vCore (at auto 1.92v input) during my testing as temps were getting a little hot before it would either hard lock or BSOD. Most BSODs where WATCHDOG_TIMEOUT errors. I was unable to get the hex code in this instance as the system would fail to write a DMP file and would just stick at 0%. It made it very tough to analyze what went wrong!

 

[Woomack]

Input voltage helps almost only when CPU passes ~4.7-4.8GHz. 4.8GHz+ in most cases already requires better cooling than standard water. I'm usually setting 2V for 5-5.1GHz for tests. If you have to raise it at so low clock as 4.4GHz then something isn't right with some other settings or motherboard is giving too low voltage at auto. At auto it's usually 0.4-0.5V higher than the CPU voltage so when you set 1.3V then it should run at about 1.7-1.8V.

One more thing is that LLC controls input voltage , not vcore. If you set high LLC and high input voltage then it will run higher than it's required.

In Win8/10 there are mainly these watchdog errors but in Win7 is usually better description like 124, 101, 50 or other code.

 

[hemirunner426]

Where would be a good place to start? LLC level 6 and 1.85v input? If the lockups continue to pop up, raise LLC level then go to input voltage if I run out of LLC leves?

 

[EarthDog]

What I would do is reset the bios to defaults and see what that value is. Set it manually, then try overclocking.

 

[hemirunner426]

What I would do is reset the bios to defaults and see what that value is. Set it manually, then try overclocking.

At default settings it defaults to auto (1.80v). When I raise the multiplier and leave input at auto it moves to 1.92v.

 

[EarthDog]

Leave it at 1.92v manually and overclock away.

 

[hemirunner426]

Leave it at 1.92v manually and overclock away.

That is exactly how I originally started out. The machine would either hard lock or BSOD with WATCHDOG_TIMEOUT during stress testing. I went the route of upping vCore up to 1.35v but it didn't seem to make a difference. I then went back to 1.30v and started raising input voltage and that's when I started noticing the machine would last a bit longer before a hard lock or BSOD.

 

[hemirunner426]

So here is what I've learned so far...

1. LLC used to control vDroop on vCore in older processors. Haswell moved the voltage regulator on-die, so LLC now controls vDroop on the VRM on-die and not the vCore.
2. LLC set to auto would add an extra .2 volts to input voltage. My actual BIOS setting was 1.98. Auto LLC bumped it up to 2.0v at full load.

I changed LLC to level 1. Now input voltage drops to 1.87v while running prime 95 27.9 custom test 8-4096 FFTs in place. I'll run this overnight to see what happens. VRM temps actually seem higher in this config as they are approaching 85C and look to be still climbing.

So what is the relationship between LLC, input voltage, and stability?

 

[hemirunner426]

Some more results:

HWBOT x265, 4k, Pmode, Overkill 4-6x seems to be a great way of sorting out Vcore/VCCIN problems. If one of the x265-x64.exe processes crashes, you know you're on the edge of stability.
So, its a matter of the process of elimination... is it vcore or vccin? Up one and test. If you seen no difference, set go back to the original settings and try the other.

I set the BIOS to 1.98 VCCIN with an LLC of level 6. This gives me a slight vdroop at load. Load vccin = 1.968. Higher LLC would cause VCCIN to fluctuate between 1.98 & 2.00v. This probably means the Auto settings uses LLC 9.

Using HWBOT x265 in the method above revealed that I had a slight instability at vcore 1.32. vcore 1.33v seems to have solved it. Once I verify this, perhaps I'll start notching down vccin and see where that takes me.

This brings me to a point though when looking at the relationship between input voltage, vcore, ring voltage... lets take a look at it from the vcore perspective. Stock voltage is vcore 1.1v or so? Stock input voltage is vccin 1.8v? We have a difference of .7v. Would it not make sense to try to maintain that relationship as you increase vcore voltage? Using this logic would put me around 2.03v vccin. Obviously its not an exact science, but something to think about.

There is really not a lot of information out there on the relationships between the various voltages from perspective of the CPU VRM.