As I keep saying too many times, I've experimented with NVMe M.2 SSD-caching for SATA SSDs and HDDs. Keep in mind that I was working more with Win 7 than Win 10 until early this month, and I can see that the newer OS has a snappier feel to it and may actually use less RAM. The experience I refer to here occurred under Win 7. I had stored my Steam games on a Barracuda spinner -- a 5,400RPM 2.5" laptop spinner!
My PrimoCache configuration including about 5GB RAM for RAM-cache, and half of a 250GB 960 EVO NVMe. Once the EVO's caching volume began to fill, I could really, really notice a difference during game-play, and it was almost scary. There was no comparison between that configuration and simply caching the spinner to RAM.
The paradox: with version 2.7 of Primo, you cannot measure the effect of the SSD-caching with conventional benchtests, but only the RAM-cache. It uses a stealthy caching strategy, filling the cache when the system is at idle so as not to hog clock-cycles and bandwidth. Version 3.0 will change that.
If I didn't believe what I saw, I wouldn't be moving cautiously toward adding a cache volume to my 960 Pro, which already contains the two OS system volumes. But I've tried it briefly so far. I just want to make sure the system is rock-solid stable, blue event logs, perfect sleep and hibernation, etc.
And that's where we are. Perfect. This will just be the cherry on top of the ice-cream soda.