Lowering voltages is an everyday thing. My Ryzen likes to run 1.475v for single core and 1.42v for multicore, so dropping VID down to 1.205v is a serious drop in not only temps but core health with high core use. Also means I go from a 41.05 multiplier across all cores to 42.8 across all cores under my specified loads. CT2 (for me) has 3 settings, idle - 18% runs stock Ryzen boosting, so simple things like websurfing etc behave just as a Ryzen should, boosting single cores to 4.4GHz. 19%-54% uses a mild setting, so I see 4.4-4.3 across any cores under use, generally around 4. Above 54%, it kicks in the more aggressive boosts and hits 42.8 solid across all cores. If I disable the P1 setting and let the cpu boost with just mild, it settles for 41.8 across all cores, which is stock settings, but at 1.2v not its auto 1.4v.
Ryzens moderate boosts according to core use, voltages and temps. But math says you need balance, so if temps go up, boosts go down etc.
Ryzens don't use a monolithic die, under the IHS is 3-6 chips, not 1. So to function, they require some sort of data path to allow communications between the various chips. This is what Infinity Fabric does. IF is based/controlled by your ram data rate. Fclock, mclock, uclock. The higher those 3, the faster the core communication, the better your performance. Left on auto works, but setting a static high value with tailored ram settings often works better.
It's like a carburetor. Perfect fuel/air ratio is 14:1. But that almost never happens. You'll generally get anywhere from 17:1 and get great gas milage and no HP, to 11:1 and be dumping gas and spinning tires. Dial in the carb for best mixture gets you the HP, but also the milage.
With ram timings, that's the amount of time it takes to open a door, enter, cross a room, open another door and exit. That's data in, through, out. The sub timings are the speed you open the door, the amount of time you wait until the door is opened, how fast you go through the door etc. Try going through the door too fast without waiting for it to open means face plant the door or stubb a toe.
So tailoring those settings and adjusting them can mean the memory controller can open, demand, get, refill data faster. Which improves performance without messing with cpu functionality. It's a freebie you setup in bios. Basically my ram came stock running 11:1 air/fuel ratio which was ok, but now I've set it for a permanent 14:1 ratio and added free HP.
Cinebench is a synthetic test. But it uses blender. A higher blender score means blender takes less time to complete the object render than a lower score. That can very much make a difference to some ppl. Rendering an hour long process in 45 minutes instead. Over the course of a day, that's more complete projects, or if folding@home etc can mean months of pc time saved.