overclocking [Discussion] CPU Voltage Tables for Turbo Frequencies? (Theory)

[CaptainXLAB]

 

I have an interesting theory about voltages in CPU Overclocking (at least I think so?)

 

Note: I have X.M.P 3200MHz Enabled for G.Skill Trident Z 16GB (2x 8GB) DDR4, seems to cause a significant increase on CPU temperatures.

 

I was trying to overclock my i7 7700k with ASUS Strix z270F.

 

I managed to reach 5 GHz but only for not-heavy loads as my CPU isn't delidded and reaches extreme temps easily at such frequencies. (had many crashes along the way too)

 

I kept trying to find a stable overclock, like from default turbo 4.5 GHz to around 4.6, 4.7 or 4.8 GHz.

 

When I tried those above 4.5 frequencies, I had to use adaptive voltage (ofcourse, keeping a high manual voltage isn't good)

 

The adaptive or offset modes stay normal when the frequency is 4.5 GHz.

 

As soon as I make it 4.7 GHz, the required voltage to stay fully stable, bumps up from 1.2v to around 1.33v (I confirmed that for my CPU using fixed manual voltage and overnight stress tests)

 

BUT, when in Windows, using Throttlestop, if the maximum frequency is 4.7 GHz and I manually force CPU Frequency down to 4.5 GHz, the voltage STILL stays too high for 4.5 GHz.

 

The adaptive voltage curve increases linearly for all turbo frequencies when I increase the voltage FOR the highest frequency. (correct me if I'm wrong )

 

Right now if I set my CPU to 5 GHz and adaptive 1.4v, it'll still reach 95C at forced manual 4.5 GHz because it's keeping much higher voltage than required by the CPU to stay stable at 4.5 GHz, which is set according to adaptive voltage curve for all turbo core ratio multipliers, when the maximum voltage value is increased.

 

Can't there be voltage tables or something? (like in GPUs?)

 

Like, for example, *I KNOW* that MY CPU stays perfectly stable at these frequencies and respective voltages and temperatures:

 

4.5 GHz 1.215v 74C Max
4.6 GHz 1.280v 78C Max
4.7 GHz 1.330v 82C Max
4.8 GHz 1.355v 85C Max
4.9 GHz 1.380v 89C Max
5.0 GHz 1.395v 95C Max

 

(just examples)

 

I want to be able to set these Frequency-Voltage values in a voltage table for better temperatures and voltages when I overclock.

 

So, if we could define voltages for every CPU frequency (for every +50 MHz or +100 MHz from the default turbo clock)

(And an automatic increasing curve for all frequencies between Default MHz and +50 or +100 MHz and similar for each next +50 or +100 MHz)

 

Currently, if my clocks are set too high, the temperatures and voltages, both, will increase a LOT at lower frequencies too, causing constant high temperatures whenever my CPU is in turbo mode.

(and with a voltage table, the primary advantage would be, that this won't happen.)

 

And I won't have to restart and reconfigure stuff in the BIOS each time I want to switch between higher clock speeds and high temps, or normal clock speeds and low temps, I'll simply have manually pre-set values in a voltage table for every +50 or +100 MHz on the adaptive CPU voltage, and click throttlestop to increase the CPU Frequency as I want.

 

This would also allow us to find the lowest voltages our CPUs can work with at every next significant frequency jump during overclocking, which would also be specific to each of our CPUs, and would have constant best temperatures if the CPU isn't under full load at the maximum frequency.

 

And, say I have achieved perfectly stable 4.8 GHz Overclock on my CPU. I want to go further high, and for that, all I'd have to do is find the voltage at which the CPU is stable for 4.9 GHz and set it in the table. If I get 85C on 4.8 GHz heavy CPU loads, and similarly 90C for 4.9 GHz, I can easily go back to 4.8 GHz when something too heavy is going on, which I know can cause my CPU to overheat, and increase the frequency again when I have not-as-heavy loads. Currently, trying this won't help at all because adaptive voltage sets too high voltage for all previous frequencies too.

 

I'm just an enthusiast and I have no idea if this idea can actually be implemented or not.

 

It's just a theory that struck my mind.

What do you all think? Would this allow much more flexibility when overclocking?