[SOLVED] Safe voltage for AMD Ryzen 9 3900X ?

[Dancingchair]

Hey guys
This is my first post here - not sure how everything works here but lets see.

So I've built a new rig with a ryzen 9 3900x and a corsair H100x - and so far, as I've been testing with ryzen master, i came stable up to 4.2 GHz and 1.30 V.
Now, this CPU stockwise has around 1.375 or 1.4 Volts sometimes 1.5 (not sure on that one tho).. which is rather high imo. But how far is it safe to push this CPU voltage without having to be scared of breaking it?

I'm rather new to this whole overclocking thing, so dont go to harsh on me ;-)

Cheers,
Dancingchair

 

[drea.drechsler]

Your question is invalid because the safe voltage for any semiconductor is dependent on a number of factors, mainly core current and temperature which is also determined by duty cycle or operating frequency. And in the case of a CPU processing load as that determines how many internal circuits (transistors) are being used simultaneously.

The CPU is designed for and perfectly safe up to 1.5V. We know this because AMD (Robert Halleck, AMD's technical marketing rep) has told us so several times. But's only safe under certain processing conditions. If you set it up right it will boost to it's rated clock (4.6Ghz) on a single core at a time in light bursty type workloads and may need up to 1.5V to do so. But as the workload sustains, or becomes less bursty, it will lower the boost clock and lower the max voltage so it might also stabilize around 1.4V. So both are perfectly safe, under the appropriate conditions.

A more valid question is this: what's the safe voltage when at maximum processing load, something AVX like CB23, workload. That would be what's called it's FIT voltage. That is variable as it's determined for each CPU during binning. It will be something around 1.25-1.325V and it won't be stable anywhere close to 4.6Ghz with it so you lose a lot of performance for the light burst processing load that's typical of gaming.

And lastly: what is 'safe' for you? The process (electron migration) is cumulative in nature and doesn't 'kill' your processor immediately. It will take a while with a higher voltage (driving higher core current and temperature) making it go sooner. So essentially, any voltage can be considered perfectly safe at 4.6Ghz if you don't mind it degrading in 6 months useage to the point it's no longer stable and has to be down-clocked to 3.8Ghz (it's base clock speed) to do anything.

BTW: The above is only relevant to electron migration processes. The other process that's totally voltage dependent is dielectric breakdown: that's when a high voltage 'punches through' dielectrics inside the CPU. Again, AMD has told us many times gen 3 CPU's are designed to be perfectly safe boosting as high as 1.5V so it seems a safe assumption that the dielectrics can withstand at least that much, and higher for safety margins.

 

[Dancingchair]

Your question is invalid because the safe voltage for any semiconductor is dependent on a number of factors, mainly core current and temperature which is also determined by duty cycle or operating frequency. And in the case of a CPU processing load as that determines how many internal circuits (transistors) are being used simultaneously.

The CPU is designed for and perfectly safe up to 1.5V. We know this because AMD (Robert Halleck, AMD's technical marketing rep) has told us so several times. But's only safe under certain processing conditions. If you set it up right it will boost to it's rated clock (4.6Ghz) on a single core at a time in light bursty type workloads and may need up to 1.5V to do so. But as the workload sustains, or becomes less bursty, it will lower the boost clock and lower the max voltage so it might also stabilize around 1.4V. So both are perfectly safe, under the appropriate conditions.

A more valid question is this: what's the safe voltage when at maximum processing load, something AVX like CB23, workload. That would be what's called it's FIT voltage. That is variable as it's determined for each CPU during binning. It will be something around 1.25-1.325V and it won't be stable anywhere close to 4.6Ghz with it so you lose a lot of performance for the light burst processing load that's typical of gaming.

Great answer! Thanks

So usually when the computer is not under a lot of load, I might see a 4.6 GHz Boost on one, maybe two cores but not more. And if the load is more consistent, as you mentioned CB23 or some other stress test, it lowers the Max. Frequency to something more sustainable.. i see, thanks!
Does this mean now that I should be able to OC my CPU until I hit around 1.5 V (Let's say I'd have the appropriate cooling for that) and accordingly i could incrementally increase the frequency with it, right?

 

[drea.drechsler]

..
Does this mean now that I should be able to OC my CPU until I hit around 1.5 V (Let's say I'd have the appropriate cooling for that) and accordingly i could incrementally increase the frequency with it, right?

Appropriate cooling to allow 1.5V at a reasonable sustained operating frequency and processing loads is going to be something like liquid nitrogen or (maybe) sub-ambient cooling solution. It's not practical but you can get very high sustained clocks that way, overclockers are getting upwards of 5Ghz and often at even higher than 1.5V on phase-change cooling even.

Whether or not that's an option depends on whether you like a refrigeration unit sitting at your feet and putting up with the maintenance the setup demands. Your motherboard and processor are sealed up under layers of silicon rubber to put off the inevitable condensation. That needs to be periodically taken apart and cleaned/checked for moisture/mold that will ruin the system. I've wondered if a solution that involves immersion in chilled Fluorinert might be viable. That way you can put the chiller in another room and pipe to Fluorinert to the tank holding your motherboard sitting on the table right next to you. RGB'g that would make an awesome display.

Ryzen's architecture seems capable of a lot more than average users can get; it's the 7nm process node that makes for difficult cooling. The die surface area is way to small to transfer heat fast enough. That's the power density problem they talk about.

I'm not a fan of overclocking gen 3 ryzen. People who do usually get performance only very slightly better than what a decent PBO can achieve for heavily loaded all-core workloads. And for that they wind up losing the max clocks the processor boosts to on single core work loads, typical of gaming. And even to get that it takes a lot of tweaking to find the balance of clocks/voltage/temperature for the workload most typical of your useage.

 

[Dancingchair]

oh damn okey^^ thanks a lot, didnt know but good to know.

thanks for your help!