My Ryzen 1700 needs 24 hours to finish each WU.
My Ryzen 3700X just completed five in 8.5 hours.

Not a good comparison tho as the Ryzen 1700 also runs 2 GPUs and is maxed out.
The 3700X only computed these five WUs, nothing else.
But still? Why is the 3700X so much quicker?

I've tried as many of the tricks as I can understand, fiddling with cache size, increasing project resource share, stopping/pausing all other projects except WCG, the "8s" timing when that was still a thing, no luck.

I'm running an i7-3770, allocating 100% of 7 cores 24/7. 16GB ram, and somewhere around 95GB drive space available.

All the fiddling I did has actually screwed up BOINC prioritizing. Before I started trying to get one of these new project WUs, BOINC would generally be running 3-4 WCG tasks, one seti@home GPU task, 2-3 rosetta tasks, and then every once in a long while a single long climateprediction would run for a week. Now, it seems to be cycling between wcg and rosetta. The log will say WCG is not the highest priority task and then get zero WUs, and run nothing but rosetta for most of a day. Then it'll grab a few WCG WUs and run nothing but them for a while, then switch back to rosetta. That's pretty annoying since it was working great beforehand, splitting the cpu time pretty evenly between the projects. I don't know how to fix this and the only thing I can think of is to uninstall BOINC completely and re-join the projects.

I even bumped WCG resource share up to a higher percentage, and it didn't help. WCG still sends only a handful of WUs and when completed it returns the message that it's not the highest priority and then doesn't send any WUs so boinc switches to rosetta exclusively.

HOWEVER, BOINC started acting weird only AFTER I had tried all those tricks on how to get one of these new project WUs.

Hey Hype! I'm running 3 systems with a Ryzen 1700, each with 16 gigs of ram. I only crunch WCG and do not crunch any GPU units (mine run as close to headless as possible). I also limit my MIP workload on each of those systems to 3 units crunching concurrently, because the 1700 only has 16MB of L3 cache, and each MIP unit uses at least 4 MB of that cache. Whenever the L3 cache is exhausted, it tends to slow down not only the MIP but all other work units. Using this system, my average crunch time for ARP units is about 16.5 hours, with a standard deviation of about 0.73 hours. More data:
Machine 1: 3 units, min 15.89, max 17.34, average 16.39 with stdev of .823
Machine 2: 7 units, min 15.82, max 18.36, average 16.70 with stdev of .889
Machine 3: 8 units, min 15.82, max 17.64, average 16.51 with stdev of .634

I configured each of the systems with different RAM speeds to see if that had any effects on crunching. Machine 1 has 2666, Machine 2 has 2400, and Machine 3 has 2133. It doesn't seem to have any significant effects on any of my crunching, so RAM speed is relegated to a footnote haha. While I'm guessing that SSD vs HDD doesn't have a huge effect, I haven't been able to test that - all 3 machines run on SSD.

Now, I don't have enough results such that n>=29, so it's not true statistical significance. But, assuming my results are not untypical, your average of 24 hours would be a statistically significant difference, being almost 10 standard deviations from the mean. So, the chance that you are just unluckily catching large or difficult units would be infinitesimal haha. My guess is that your other work units are having a knock-on effect, or your GPU units are leaching cycles from your other work units causing a general slowdown. Between those effects and the year over year IPC improvements, and clock speed increases, I'm not surprised that the new silicon is significantly faster on these units. 8.5 hours is blazing fast - very impressive! That said, 24h results on your older silicon aren't terrible results, and may be a fine compromise depending on your goals for other projects

I wish someone with powers or skills could/would fix the heading of this thread

Looks like the work has dried up again.