Power / Current Limit Throttling - what to try?

[telliott08]

I'm new to overclocking.  Have been using HWMonitor and XTU2 to try and get my new rig set up.

My overclocking goals are:

1. Don't let the smoke out
2. Get something close to maximized, stable performance that won't adversely impact the life of my gear.

When I started, I was getting a LOT of thermal throttling.  I think I have that pretty well under control now, no temp throttling, cores sit at about 98c when at 100% utilization for extended periods.

What I am seeing is Power Limit Throttling and Current/EDP Limit throttling while running benchmarks.

(Cinebench R23 33503 / XTU2 11065)

If I understand it right, the CPU would like more volts, is the correct?

Regardless if I am right or wrong on the voltage - What can I do to try and resolve this?

Guessing I need to do something in the bios but have no idea what...

Hardware:

i9-13900K | ASrock z790 Taichi | 32GB DDR5 6000 | Corsair HX1000 PSU | RTX4080 | 360 H2O

 

TIA for any assistance!

[telliott08]

Some screenshots while running Cinebench multi for a few minutes in hopes it's informative:

[Mr.Scott]

Temp too hot. You're showing 110c average.

[telliott08]

Yeah, I had been playing around with some of the power profiles in the BIOS to try to address the throttling.

Here's a screen shot after I reverted the power profile back while running Cinebench:

is this still too high? 

Any thoughts on the throttling issue?

Thank you!

[Mr.Scott]

Still too hot. Temp throttling decreases power to try to get temp under control.

Lower your overclock and voltage and try again.

[wutske]

Apart from the high temperatures, the power limit actually means it's drawing more power than it likes to, so adding more volts will make this worse. Adding more volts also makes it run a lot hotter and 13900K are already incredibly hot and hard to cool. So I would suggest finding an overclock that works at lower voltages. With a 13900K, you can also achieve beter performance by undervolting it so that it power consumption lowers and so that it can turbo for a longer time.

Not much experience with XTU, but there should probably a setting to change the turbo power limits (PL1/PL2), though at these temps, I think it's best to not touch it since your cooler already can't handle the power output.

[telliott08]

Thank  you for the feedback - very helpful.  Seems like the more I get into this, the more questions I have...

I played around with the PL1/PL2 limits (Turbo boost power max / Turbo boost short power max in XTU).

Even if I drop them both to 85w, I still get Power Limit Throttling (although my temps drop WAY down - as does the performance....)

I set them both to 270w for now as it seems a reasonable compromise between performance loss and high temps - was kind of targeting 90-95c under sustained load as, if I understand it correctly the processor temp throttles at 100c but is supposed to be comfortable running below that.  Please let me know if these temps are too high still - and what I should be targeting if my assumptions are off...

Screenshot below, I am running folding@home on the max settings as it's about the worse case scenario for processing I can think of.  It pegs the CPU & GPU at or very close to 100%, so generating the most thermals and power draw... 

What I notice is that although I set PL1 and PL2 to 270w, the actual package draw is slightly higher.  I am guessing that's why the power limit is kicking in... 

Is there some other setting I should work with to prevent this?

As for undervolting, I have run into something of an issue where the BIOS has a setting for "Undervolt Protection" which is enabled by default.  This setting prevents me from using a negative offset in the software (XTU, ThrottleStop, etc).  Voltage values in the BIOS currently are set to Auto.  I have tried using a negative offset down to the max the BIOS will allow (-100mv), but have not seen any change in the power limit throttling behavior, so I reverted back to the Auto settings...  I also did not see any change in the voltage values while running cinebench etc in HWMonitor, which makes me think that my changes to the offset are being ignored...

 

 Please let me know what I might want to try for next steps & thanks again!

 

[wutske]

If you set the power limit to 85W then you're pretty much going to hit the power limit throttling all the time.

PL1 and PL2 are values that define how much power your CPU can draw, so if you limit it to 85W then you're setting a much stricter limit. PL2 is usually a higher number and pretty much defines how much and how long (Tau) your CPU can throttle. So if Tau is set to 20s and 250W, then you're CPU will be consuming roughly 250W for a maximum of 20s before it's throttled back to PL1. If you have sufficient cooling capacity, setting PL2 and PL1 to very high numbers (or unlimited/unconstrained) then you won't hit the power limit anymore. This is a good article: https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo

The TDP is not entirely the same as your CPU's power consumption, in fact, it's usually lower since TDP defines the amount of heat the CPU will generate, but heat is just a byproduct of work being done.

 

Now, how does undervolting allow you to go much faster; the power consumption of your CPU is defined by two values: the VCore and the frequency at which it's running (I'm just going to ignore the GPU and memory controller and also the fact that temperatures also have a play in this). The higher the VCore, the more heat it generates. The higher the frequency at which it runs, the more heat it generates.

Let's say we have a completely imaginary example, your 13900K at 1,3V running at 5,5GHz will consume 250W of power, which is the PL2 I mentioned before. Let's say we drop the VCore to 1,2V and keep the frequency at 5,5GHz, the power consumption drops to 220W (again, I make these numbers up on the spot) but your PL2 allows your CPU to consume 250W. This means your CPU can boost to higher clocks, let's say 5,7GHz, et voila, free performance ?.

So basically, by dropping the VCore, you can actually have a CPU that can run faster (as long as it's stable at that VCore of course).

 

Obviously, finding the right VCore at which you can maximize the maximum frequency is not easy and it'll take a lot of time to get to the sweet spot.

 

//edit: about the temperatures: 90°c nowadays is kind of okay-ish. It's not great for the silicon of course when you run it like that 24/7 but a normal desktop isn't running at full load 24/7 either. Not sure if it's summer where you are, but keep in mind that in the summer your room temp might be 10°c higher so it might still throttle on hot summer days.

Edited February 15, 2023 by wutske