OK, here's my data so far, but with the usual preamble that it's data collected using my specific environment over the years (though when I upgrade I go back and re-benchmark the earlier versions). In that context, it's actually pretty meaningless other than as an indicator as to where
I'm seeing performance improvements or degradation. All I can say is that my own results seem to confirm that many improvements have been made as advertised, but it doesn't surprise me that not everyone is seeing the same improvements as
@Karayuschij's linked post from the U2U above would confirm. Why that user is getting poor performance is impossible to say, my only observation from a brief glance is that I'd prefer to see larger sample sizes, but that still might not matter of course.
Anyway, OSX first:
All looks very positive, standard previews improvements especially so. The one "slower" category (exports of edited files) can be ignored as it's a tiny difference and I also know that no performance changes were made in that area.
Next, Win10. These are split into two sections, one where Hyper-Threading (HT) is enabled (probably the default for HT-capable systems), the second with HT disabled:
With HT enabled, results look OK. Not as big an improvement as under OSX, but healthy enough for all that. Standard previews, for some reason, seem blisteringly fast, even faster that OSX.
But note the asterisk against the three tasks on "edited" files. What I was finding when using Win10, and more so with Classic than previous versions, was a definite slowdown the longer the task was running, e.g. the more images in the sample size, the larger the per image time would likely be. Normally, my testing routine would be reboot, open Lightroom, run task 1 for both sets of images consecutively, then reboot, start again with the next task. Using this normal method (which is how I do it under OSX as well) I was getting much longer second runs times per task than I expected, so I modified the approach to do a simple restart of Lightroom between Set1 and Set2, and the "*" indicates that the specific test was following a relaunch. This information has been passed to Adobe, so I'm hoping to see this area improved as the engineers continue to work on Classic performance.
Now, Win10 with HT disabled:
Still better than LR6, though it could hardly be worse. I'm not sure how much effort, if any, the engineers are putting into optimising on HT-disabled systems, but it's not really coming through. Some of that (maybe all of it) is likely explained by the "parallelisation" of these tasks now, thus the more logical cores the more concurrent previews can be built. That's a large part of the significant step forward under OSX (8 logical cores) and Win10 with HT (again 8 logical cores).
I've probably got some import timings documented somewhere, if I find them I'll post in an appropriate place.
Just a final word to reiterate that caution needs to be used when examining (or broadcasting) this data. It's not the "gospel according to Jim Wilde", more a (hopefully helpful) indicator that there have been performance changes incorporated into Classic, though it's a pity if not everyone sees them. But's that's been the nature of "Lightroom's performance problems" for years, some tear their hair out in frustration, others wonder what all the fuss is about. I doubt that will ever totally change, though it has been good to see some positive reports on the forum.