Help with undervolting?

[marcv81]

Hi team,

I'm in the process of undervolting an i7-8700 (non-K) in a Dan A4 SFX. I found a voltage I cannot go below, and I'm now reducing the offset back to a level where the system is stable. I run Prime95 but it does not find any error and the system freezes instead. Below are my times to freeze so far.

- 140mV: 1 minute
- 135mv: 10 minutes
- 130mV: 25 minutes
- 125mV: 25 minutes
- 120mV: 9 hours

Would you say Prime95 is a good test, or is there anything else I should try? So far the freezes happen when I use the system, not when I leave Prime95 alone. I'm guessing user interactions might cause voltage fluctuations (e.g.: the WiFi chip waking up) that are more relevant to stability than a high CPU load. Any idea?

Once I find an offset with no freeze/crash for 24h, would you know how much safety margin to add? Would 10mV be enough, or should I maybe go for 20mV? Stability is a priority: I sometimes leave the system to crunch numbers overnight and I can't just reset it when it crashes.

Thanks!

 

[bichael]

I would just set it at -100mV and see if it's stable overtime. I always used x264, if it ran okay for one hour then okay, back off a little and see how it goes. Unless you are very thermally / power limited I wouldn't stress about trying to dial in the absolute lowest voltage.

 

[NateDawg72]

Prime95 is a good test. I've also used LinX, and I've head good things about Asus' RealBench if you'd like to test several.

I think you're on the right track - I'd suggest finding an offset with no freezing / crashing for at least 24h, and then add 10mv to that.

I don't require the same level of stability for my machine, but I've always also factored in a day of heavy "normal" use after my undervolting - every once in a great while I've had a program or game that manages to cause a crash/freeze that the stress tests didn't.

 

[marcv81]

Thanks for the helpful replies @bichael and @NateDawg72. It survived 18h at 115mV through a combination of stress tests, gaming, and browsing. I happen to have time now where stability is not critical, so I will keep backing off progressively. But I think both your replies are spot on and it'll likely end up somewhere near 100mV, 10mV lower than the 24h stability point.