I do it for a hobby really. Love the challenge.
I do a LOT of tests! I'm focusing on 3600 currently as I've had it running stable, and better than my 3533 profile in all tests. Not much, but enough to be noticeable.
After many tests today on 1103, I cannot get my profile to run reliably as it does on 1002 - I've loosened up all my secondary timings, changed voltages up and down, but nothing works. It's intriguing and annoying! - I thrive on trying to find out what's changed (AMD tell us nothing for a multitude of reasons) - if I can.
Since I started with My X370 Prime Pro I've gradually got better and better at doing things, and been able to extract excellent performance. But, give a dog a bone (or in this case, 4200MTs ram) and he's going to do his darnedest to get every bit of juice out of it I can! I should point out I have ADHD, and can do this for HOURS! I do lose hope sometimes and give it a break - often coming back too it able to solve my current problem.
At one point, I had 3600MTs stable with 1.420v, now (if I'm on 1002 bios) it needs 1.30v. Any setting either side of those voltages would cause failure (for me and my set up) - it's got to be spot on, so takes some time working in combination with other voltages, power settings.
I may now have found the limit for my hardware, but it will take a bit more testing to be sure of that. I think on 1002 I can probably tune in 3666, but not on newer bioses. It may be that LLC is working differently on the latest bios, or it may be I need to change the termination blocks which I've not previously needed to do. Really am stumped why previous timings and voltages no longer work.
Something else I've noticed - for my setup anyway - is that there are some tweaks from the ram calculator which really helped out - as well as a couple that didn't. L1 and L2 Stream HW Prefetcher>Enabled worked well and Opcache>enabled too, but enabling mem interleaving at the suggested values actually SLOWED down certain tasks. These are settings that may bring benefit at other settings for my kit, but I do those tests last.
You're dead right about OCing being able to make things worse... One thing I found was that if your memory OC is not quite right your machine will take some time to boot. When I zero in on good settings, it's often the thing I notice first - the machine booting at a decent, normal speed, as it should. Another thing to know is that sometimes loosening (not tightening as you may think) a timing, can increase performance. This is because bad OCs can slow down your hardware as it's not running optimally.
I do try to repeat test at least 3 times - when I have a stable OC I will repeat them again. This way you can average the performance increase/decrease. Not perfect, but works for me.
I have lots of time on my hands lol!!
Keeps me occupied while off work and waiting for surgery, but i've always loved this kind of thing right from the BBC micro days!