Everything related to Zen 2 automated overuls is related to two important aspects PBO and AutoOC. Depending on AMD, we can both improve up to a point 200 MHz on a single basis with constant sustained frequency across cores.
This is of course part, but elsewhere it is not, where in addition to this article we have to add another ingredient to the voltage-like equation of the top series that includes all processors. Ryzen 3000, either inactive or full.
What should be considered is to increase the frequency of reinforcement
We've talked about this a lot since August, so we'll summarize it a lot in a logical way: SoC Power (PPT), Current VRM (TDC limit) and temperature (XFR 2). AMD uses its Precision Boost 2 algorithm in the order of these three components, where if the CPU sensors detect that they are not close to any of those parameters, the algorithm will increase the clock speed in all possible directions.
With this in mind and understanding that PBO and Precision Boost are one technology and algorithms that are compatible with each other, what happens if we don't use PBO and reduce the three parameter levels in which Precision Boost 2 is dominated?
That we get a different situation from what may be excessive and with a pure mind, the speed must be equal. How can this be? Following the guidelines outlined and using the analogy of the triangle set by AMD, when moving away from the three sides of the figure, the algorithm should find a better shape to increase the speed.
The most important question, then, is how. Well, it's all related to a simple parameter: voltage. What we will do is very simple and efficient and is nothing more than a powerless control over two CPU voltages: Normal electrical power and BMI voltage.
Decrease voltage to increase speed over PBO?
This is exactly what is achieved, a very high speed thanks to low volume. At low voltage SoC power extends from its threshold, the current VRM does the same by reducing CPU usage and therefore MOSFETs deliver less amps and pure logic, lower voltage equals lower temperature, So it's all aligned us.
To accomplish this, we will need to go to our UEFI and do a undervolt For both CPU Voltage and IMC Voltage, you remember that the effect will be very high for every millivolt we drop. That is, lower voltages will mean higher frequency, so we will have to test where both converge.
Let's look at a practical example:
Depending on trial and error we set the voltage a negative decrease of 0.18875 CPU volumes and 0.05 volts of BMI. We confirmed this solidity during the 8 hrs of the Rebenbench, after which and after allowing the water of our RL to cool, we continued the associated testing.
First, we use the Cinebench R20 for tests where we set up the usage and usage of the CPU and Ryzen Master.
As you can see from this stock optimization, CPU usage is about 70 watts on average, which affected frequency (4,114 MHz) and electricity (1.38V), resulting in the 66.76 degrees based.
Activating PBO usage has been 74W, the temperature subtends 3 degrees, the frequency goes up 4,124 MHz and the electricity did the same and went up 1.41V
Finally, we have our fix with Underclock, where the Ryzen Master dislikes the voltage well (something the CPU-Z does) but it looks better as usage is down 57 watts, the temperature is maintained there 60 degrees it usually goes up to him 4,152 MHz.
You can also see how the SoC and TDC powers pass from 13W and 66% at 11W and 58%, a significant reduction in overall goals is undoubtedly driving the final frequency and with this we achieve greater performance without the need for any form of overlap in BCLK or multiplication.
This at the same time means that AMD is closely monitored. A full ten-fold drop with Undervolt is not the most common in previous builds, but it does show the good work of the technology they work on.
