Overclocking - 'all core' vs 'by core usage' for daily use?

[Kalm_Traveler]

Hey guys this is something that I've never seen a discussion on so I'm curious what the community thought is.

 

Nearly all the overclocking guides, articles, videos etc around are all about 'max stable all-core overclock' settings which makes sense for highest benchmark scores etc, and even back when we were only gaming on 2, 4 or even 6 core CPUs.

 

With my own 16 core rig, I do the usual max-all-core OC only for outdoor winter benchmarking partly because this 7960x can pull over 400w by itself at 5.2ghz all core and partly because I don't see a need to run 16 cores screaming fast when the majority of the time I don't hit more than 8-10 of them at the same time. 

 

Because of that, for daily use I just do a 'by core usage' OC, that way only the cores that are working will boost up - less power draw, less heat etc means (I think) stable and relatively quiet.

 

Now that the mainstream is starting to push 8-16 cores, do you think that a 'by core usage' OC for daily use makes more sense since even games are still not generally hammering that many cores together at once?

[Fasauceome]

7 minutes ago, Kalm_Traveler1 said:

Now that the mainstream is starting to push 8-16 cores, do you think that a 'by core usage' OC for daily use makes more sense since even games are still not generally hammering that many cores together at once?

yes, I personally tune overclocks to per core instead of just slapping an all core on it, because the granular nature is pretty handy and also more efficient. If I'm playing a game that only uses 6 threads on my 12 thread i7 I want to squeeze a little extra MHz out of it. I can also rely on some lower voltage to get the performance I want, as perhaps I can set a decent speed for most cores and leave the all core value an unimpressive speed for those rare occasions when I hammer the whole CPU with a workload, so I can tune the voltage better.

[Jurrunio]

Problem is "how do you stress test". If you dont reduce voltage the efficiency gain is small, but if you do there's no way to reliably stress test them.

[Kalm_Traveler]

6 minutes ago, Jurrunio said:

Problem is "how do you stress test". If you dont reduce voltage the efficiency gain is small, but if you do there's no way to reliably stress test them.

my 'stress test' is just using it. If it crashes I'll ever-so-slightly increase the voltage and try again. 
 

Right now on my 7960x I have it only doing 4.7ghz 'by core usage' with 1.2v because it never gets above 40c with normal use. Cinebench R20 crashes, but it never has any issues at all with my normal daily use (homework, gaming, running some VMs - often all at once).