First Pass Overclocking Results i9-10940x on Rampage VI Extreme Encore

[rosefire]

I would love it if someone could explain why the ASUS 0603 BIOS Optimized Default Vcore is set to the ridiculously high voltage of 1.237V. I found this voltage was about what is needed to be all cores stable at 4.8GHz.

My first goal experimenting was to find the lowest voltage my CPU would run at using stock core ratios. This voltage would be the first value in a table of core ratios and the the corresponding minimum stable voltages.

STOCK CORE SPEEDS - DEFAULT VOLTAGES
>>>>> Settings: Adaptive Mode, Vcore = 1.237V, offset = 0.0V
>>>>> Result: 0.791V idling, 1.237V full load, TPD = 190 Watts, Vcore = 45C, Vpackage = 60C maximum peak

I have no idea what ASUS Levels mean IRL, but all of the following tests were done with the default Level 2.

Starting at 1.12V, I began incrementally reducing the core voltage looking for the lowest stable voltage, but instead hit a floor where reducing the core voltage and offsets further stopped having any effect. This occurred at Vcore =1.08, offset of 0.0V, which produced that actual voltages .731V when idling, and 1.057V when stress testing.

Perhaps someone can explain where the voltage floor comes from, and why the Vcore of 1.08V results in 1.057 rather than 1.08V.

STOCK CORE SPEEDS - MINIMUM VOLTAGES
>>>>> Settings: Adaptive Mode, Vcore = 1.12V, offset = 0V
>>>>> Result 0.731V idling, 1.057V stress test load, TPD = 168 Watts, Vcore = 43C, Vpackage = 57C maximum peak

Seeking to understand the difference between the settings and actual floor voltages in adaptive mode, I probed for a lower limit static mode. Here's what I got:
* 1.10V resulted in 1.10V actual - successfully passed short duration stress testing.
* 1.09V resulted in 1.08V actual - successfully passed short duration stress testing.
* etc, on down until...
* 1.05V resulted in 1.05V actual - fail stress testing.

The lowest stable static mode voltage was found the first step that went below the adaptive voltage floor of 1.057V.

Now it was time to check out the 4.8GHz all cores clocking most i9-109xxxx chips seem to be achieving (or at least posting). Here's the result.

4.8GHz ALL CORES STABLE (reported speed 4.79GHz)
>>>>> Settings: Static Mode, Vcore = 1.180V <---IN LATER STRESS TESTING I HAD TO UP THIS TO 2.20V FOR STABILITY
>>>>> Result: 1.179V stress test load, TPD = 233 Watts, Vcore = 53C, Vpackage = 68C maximum peak

This made me happy because the 1.18V required to be stable at 4.8GHz is very reasonable. On the other hand, 68C is getting pretty hot for use 24/7.

I have not fully tested the 5GHz all cores overclock (which results in 4.99GHz reported), but it was short-term stress testing stable and the numbers looked reasonable at Vcore=1.275V. Nonetheless, the 24/7 sweet spot will likely be 4.8GHz, and maybe 4.9GHz.

I believe this chip will run all cores at 5.1GHz with a Vcore a bit over 1.3V. The thermal trajectory for the package at 5.1GHz would be around 75C, which is too high for 24/7. The ability of my loop to remove the heat ultimately be the limiting factor.

-------
Notes: Testing was done with single loop CPU-only water cooling 2.8L/minute, 27C ambient, water temperatures long-term stable at 32C in, and 33C out. The rig hardware is described in my signature. Voltages were set using XTU after booting with Optimized Default BIOS settings. Voltages, multipliers, and clock speeds were acquired using Asus Intel Extreme Tuning Utility (XTU) and AI Suite. Core temperatures were from the MB OLED panel.

[rosefire]

My first post described the results of my initial short term testing. With further testing I've found the 14-core i9-10940x to be very different from my Rampage IV Black, Edition with its 6-core i7-4930K. Two major differences:

1) Temperatures and TPD peaking near 90C and 300W respectively are limiting the stable all-core clock of my I9-10940x to 4.8GHz at a modest 1.25V. In contrast my i7-4930K hovers around 65C at 4.75GHz with an aging AIO cooler, with a further multiplier increase requiring a substantial voltage increase to near 1.5V, a level I 'm not comfortable using.

2) I'm finding that there is a very large voltage difference between short- and long-term stable voltages for the i9-10940x. In my previous post I reported finding 4.8GHz short-term stable at 1.080V, but I've found 4.8GHz is still not long-term stable at 1.245V . My earlier testing was done with XTU's linpack stress test. I've been surprised to find that CineBench alone pushes core temperatures some 5C higher than XTU. But because XTU filles in the brief drop in loads between CineBench iterations, running both XTU and CInebench simultaneous requires still higher voltages to stave off computational errors.

At this point, long-term solid stability on all-cores at 4.7GHz looks like this:
VCore: Adaptive mode with a -.096V offset results in a full load voltage of 1.085V.
Peak Core Temperature (XTU and CineBench at the same time): 69C
Package Temperature (XTU and CineBench at the same time): 79C, Peak 82
Peak Power Dissipation 299W
Cache Ratio: increased from 24x to 28x (increases benchmark performance 6.4% )
Cache Voltage: Adaptive Mode with a -.060V offset (only reduces temperatures a degree or so)
No memory overclock attempted yet.

Environment:
Ambient Temperature: 27C
Loop Water Temperature (sustained): 32C inflow, 34C Outflow

Next step is to try optimizing core temperatures and voltages independently. If there is interest, I'll add more to this thread. I also will soon have a second i9-10940x to play with.


If you find this information useful, have anything to add, or disagree with my conclusions, please post a reply. If there is interest I'll probably be inspired to post more.

[rosefire]

Measurement and graph below.

[Cezarycezary777]

Hi
What setup did you use for AVX anD AVX512 instruction?

[SilentOption]

Is the 64 C under full load or idle

[Cezarycezary777]

No, under load is going up to 95.