Thoughts about LLC with ryzen overclocking

[btmedic04]

I recently updated the bios on my Asus Strix X370-F from the September build to the most recent and have been toying around with my overclock. With the September bios my R7 1700 needed 1.3875v with LLC level 5 to run stable at 3.8ghz with my ram at DDR4 3200. With the new bios, I was able to run 1.3875v with LLC level 1 @ 3.8ghz /DDR4 3200 Prime95 stable. During stress testing (realbench and Prime95 26.6) I noticed my voltage dropped to 1.325v as reported in HWinfo which got me thinking about setting my LLC to 5 and set voltage to 1.325v and test for stability (currently about an hour into it now.) Im just curious to get a general consensus on load line calibration on ryzen. Is running a higher level of LLC with a lower voltage better or worse in terms of longevity compared to a lower level of LLC with a higher voltage?

 

[MeanMachine41]

LLC is there to provide stable voltage when demand transitions to high and to combat Voltage droop.
Usually combined with CPU Current capability, I have found a mid range LLC to be of benefit when Overclocking.
It also depends on the PSU and it's efficiency too.

LLC varies with differing systems and MBs. Trial and error gets the sweat spot. Excessive LLC can Overvolt and create more heat.

Longevity of the CPU is keeping Core voltage to min and within acceptable thermal margins.

 

[btmedic04]

Im not sure how quickly HWinfo can read the voltage sensors but I havent seen my voltage jump any higher than what I set (1.325v) but I've read that it can jump really high for a split second with high levels of LLC. I really wish I had a voltage contact pads on my motherboard so I could check As far as temperature goes, 58c for in place large FFTs, 68c for in place small FFTs. I think ill let in place small FFTs run till I get up tomorrow and see if its stable, then play around and see if I can lower LLC to level 3 to see if that reduces temps a bit while still maintaining stability. Also on the to-do list is see if I can get a little more clock speed out of it since I've been able to lower my voltage with the new bios

 

[MeanMachine41]

It's up to you ofc how you stress your system for stability however unattended runs with P95 I don't recommend. If during the run a worker stops then you've got nowhere as it will not identify why. Just that one of your Cores failed.
If temperatures rise to almost thermal margin max and your not there to stop it. Why would you want to do that. P95 will push your CPU to the MAX.

There are much better stress testers IMO, AIDA64 is not so stressful and adequate to check stability and it provides for testing your subsystems.

If you want a realworld tester then use ROG REALBENCH. If you can pass that benchmark then you are 24/7 stable.

Temperature sensor and voltage readouts are best done using HWinfo64 which will accurately read Tdie.

 

[btmedic04]

hehe I learned to overclock during the Athlon 64 days, so 12-24 hours of Prime95 is a carry over from that era. If it can pass the absolute worst case scenario, then I consider it stable. Realbench is great too and I usually run the stress test with 16gb of ram usage for an hour before starting my Prime95 run.

As for an update, my new overclock did pass its 12 hour Prime95 small in place FFTs test with LLC set to 5 and voltage at 1.325. Temps topped out at 69C. I now have a little bit of voltage and thermal head room left to play around with though to be honest, I think 3.8 will continue to be my daily settings

 

[MeanMachine41]

Well that's great btmedic04 Enjoy your system.