Question Can undervolting stop making sense past a certain point?

[rbogomolec]

I've undervolted my i5 9400f by -125mV and the results were great. Demanding games went from terrible fps drops to stable 60 fps, CPU temperature significantly fell and even the high CPU usage % went down. CPU still running at max MHz, so no loses there. System is stable, no crashes.

However, I kept on going and even reached -150mv (CPU running at 0.9998 - 1.0046 V now). Thought the system is still stable, I stopped seeing any improvement in performance. Even the temps stayed exactly the same as on -125mv. Since -0.25mV is a significant drop (at least I think so, maybe it's not really) I expected to see further improvements.

So, my questions are: Does it make sense to undervolt even further? And SHOULD I go further or am I already hitting the general limit for this CPU? By general limit I mean that I know about the silicon lottery thing and that no "same" CPU is exact the same, but I don't think many people went below 1V on this CPU.

 

[Lutfij]

To answer your thread's question, yes, it can become pointless.

You will see BSoD's/instabilities due to lack of power towards components that need said power that's being cut off during a workload. If you're seeing the same results at lower voltages, stick to the lower voltages though you should keep in mind that a couple of hours or reboots later, you could very well face instability issues.

I'd stick to the -125mv that you've reached.

One more thin, what aspect of the undervolt did you touch? There's Vcore, cache and iGPU in case you didn't know the others + in case you're using ThrottleStop.

 

[rbogomolec]

To answer your thread's question, yes, it can become pointless.

You will see BSoD's/instabilities due to lack of power towards components that need said power that's being cut off during a workload. If you're seeing the same results at lower voltages, stick to the lower voltages though you should keep in mind that a couple of hours or reboots later, you could very well face instability issues.

I'd stick to the -125mv that you've reached.

One more thin, what aspect of the undervolt did you touch? There's Vcore, cache and iGPU in case you didn't know the others + in case you're using ThrottleStop.

Oh sorry, I forgot to write that... Yeah, I'm using ThrottleStop. I did CPU core and CPU cache both by the same amount. I didn't touch the iGPU setting cuz this CPU doesn't have an integrated GPU. As far as I know, that's all that I should do when undervolting. Did I miss some tweaks, or is core and cache all I need to tweak?

-140mV also already proved to be stable (multiple reboots and gaming sessions went without a problem). I did put -125mV and -150mV back to back now in Resident Evil 4 Remake, in the same crowded fight scene. I think that there was a 1-2'C temp drop after all. I think I'll stick with -140mV for now, -150mV does seem a bit stretched already and the gains are getting quite small after that big breakthrough that I had with -125mV.

 

[uWebb429]

core and cache

The core and cache are the two most important items to undervolt. Any of the other ones can cause instability before you see any significant drop in power consumption so I never bother with them.

Do some Cinebench testing. With the cache held at -125 mV, some people continue to see improvements in performance or temperatures with the core set to -150 mV, -175 mV or -200 mV. The core and cache offset voltages do not need to be set equal to each other.

If you set the core and cache both to -150 mV, you are probably already over the edge of 100% stability. Just because a computer does not crash during a game does not mean that it is 100% stable. Some minor errors will start to happen before a CPU actually crashes. Try running the TS Bench 960M test. It is good at reporting errors before a CPU crashes when an undervolt is too much.

There is no reason to try and get the voltage to the absolute bare minimum. It is always a good idea to leave a little bit of extra voltage to guarantee that your undervolt is 100% stable.

 

[drea.drechsler]

...So, my questions are: Does it make sense to undervolt even further?....

Undervolting as far as you can is almost always better for the device for a variety of reasons. For instance it reduces heat stress on the part, it lowers energy useage and there's something called Time Dependent Dielectric Breakdown (TDDB) too. That's where any semiconductor breaks down it's dielectric when a voltage is applied over time, the lower the voltage the slower the rate and longer it takes for it to make the part unuseable.

Every device manufactured is uniquely capable of operating at a different voltage but manufacturers don't have the time to find out the limits for each individual part produced. So they use a pretty large margin to assure stability for the end user. They have ways to calculate the TDDB, though, so that they can be assured whatever they spec can allow for a decent service life. But there's almost always considerable room before you get to the practical limits of stability of your part, just take the time to test before you declare it good

And that's the point it becomes counterproductive, the point where stability suffers or other problems with performance. That's the practical limit to how far you can take it.

 

[BFG-9000]

Undervolting makes sense if it is thermally limited and you cannot change the thermal solution such as in a laptop. 9400f is a desktop chip, and in a desktop you are better off changing to a better cooler instead.

Modern processors have almost all of the headroom extracted from them already, and by undervolting you are using what little has been left by Intel for adverse conditions such as high environmental temperature, or to cover gradual degradation. Meaning that it may well work under the conditions you are using it for now, but you must be vigilant in regular stability testing in case it starts to no longer do so, or else creeping instability can corrupt your data or the OS.

One of the few useful things that Microsoft's telemetry has provided us is the proof that the error rate of PCs is much higher than previously believed, and they show that underclocking from stock considerably reduces errors. Presumably overvolting can as well, but make no mistake, undervolting is clearly overclocking.

You've got to ask yourself one question: Do I feel lucky? Well, do ya, punk?”
--Dirty Harry

 

[drea.drechsler]

Undervolting makes sense if it is thermally limited and you cannot change the thermal solution such as in a laptop. 9400f is a desktop chip, and in a desktop you are better off changing to a better cooler instead.
...

I don't see that as an either-or proposition...why not do both?

...undervolting is clearly overclocking.

I think that depends on how you define what "overclocking" is.

Looked at in the traditional sense overclocking (alone) raises clocks to exploit design margin. That increases stresses on the part which adversely affects FIT. But undervolting alone lowers voltage to exploit design margin which lowers all stresses: heat, voltage and core currents are all lowered which significantly improves FIT. That's all at the same clocks as previous so I can't see how that's overclocking

But in the modern sense, I see where it's more complicated. That because today's boost algorithms might be working at odds: the boost algorithm may see the lowered heat/current/power as headroom to allow higher or longer boosts which raises stresses again so that may be considered overclocking. But how that affects FIT is really unclear even though I believe the net effect is for the better.

But then, there is what the part manufacturer calls "overclocking". They could include anything they want in the term so read the fine print in the warranty documents to ferret out the details on that.

 

[TerryLaze]

Looked at in the traditional sense overclocking (alone) raises clocks to exploit design margin. That increases stresses on the part which adversely affects FIT. But undervolting alone lowers voltage to exploit design margin which lowers all stresses: heat, voltage and core currents are all lowered which significantly improves FIT. That's all at the same clocks as previous so I can't see how that's overclocking

With most current mobos coming with settings that has all the auto turbo-boosts set to 11 undervolting does increase the clocks since the temperature gets lowered so there is new temp headroom to exploit.

 

[drea.drechsler]

With most current mobos coming with settings that has all the auto turbo-boosts set to 11 undervolting does increase the clocks since the temperature gets lowered so there is new temp headroom to exploit.

Yes..I noted that about the temp headroom too...and that I tend to think the net effect on FIT is for the good and that's what matters with respect to concerns about a processor's service life. But it would also suggest the margin for an undervolt (until it goes unstable) is not going to be as much as on another board that doesn't do that. That could make people erroneously believe their processor is a poor sample.

But the weird things mfr's are doing with their motherboard presets is pretty scary all on it's own, and that extends far beyond things like processor boost parameters.