Mta88, very nice trick! Though I believe it considerably increases your starting time
And I definitely believe that the improvement of the latency made a big difference for you - it has the same effect on me.
I presume you run run German or Vintage D under Kontakt. If so, do you have its preload buffer turned on? This buffering eliminates all traces of latency attributed to disk access.
If it was turned off, then latency would be bad ... and then switching to RAM disk would produce just about the same result as enabling the preload buffer.
MacMacMac, I have a completely different experience with this.
First, you can't turn off the preload buffer. You can only override its default size, which is 240kB.
Second, in my case, increasing the preload buffer actually increases the minimum latency I can achieve. I believe that it works like this: This preload buffer size not only denotes the size of "preloaded" part of each sample, but also the size of chunks by which the samples are dynamically loaded. And it seems the loading, decompressing and processing larger chunks takes more time, therefore, in my case, it all runs more smoothly if it's done by smaller chunks. Of course, other peoples' mileage can vary.
Someone somewhere once said (I think it was on the Pianoteq forum) that the latency of a real acoustic is 30 ms. Not sure how true that is, but there you go.
FWIW, member Voxpops once in this
thread reported on various interesting papers regarding piano function and modelling. In one of the papers (I believe it's the first link - called 'this paper') they measured the time between key bottoming and hammer-string contact. It was between +2ms (forte stroke) and -12ms (piano stroke). Of course, this number is not directly comparable to what we call latency, but I still think that 30ms is a strong overestimation of the real latency.