...
UPDATE: in a fit of pique I deleted the amd.txt file for both rigs and let xmr-stak automatically generate them. Hashrate is depressingly low - 1020-1060H/s for the RX 570 + Ryzen 5 1600 system (previously ~1180H/s), and <1600H/s for the 4x RX 560 system (previously ~1900H/s), but at least it isn't crashing anymore (and I hope I didn't just jinx myself...).
yeah with the upgrade, all the .txt files need to be re-generated (has something to do with formatting and variable fillout for the new variables stored in the file).
...
Hashrate seems to be minus 1/4-1/3 or so compared to cnv7 (accounting for my 470 when it was plugged in as well). So far, no issues with the rx560.
I'm still trying to dial everything in with respect to stability and hashrate but as of now I have gone back to 2 threads on the RX 560 rig, using all the other auto-generated parameters except reducing the intensity to less than half what was specified for 1 thread operation (e.g. - 768 for 1 thread goes to 376 for 2 threads). Right now the 4x rig is doing a little over 1800H/s so a massive improvement from the <1600H/s I was getting before! However, I just made this change so now I need to see if it remains stable as I was still having problems with the 1 thread/768 hashrate config (e.g. - hashrate dropped from 400 to 300 on one of the cards sometime over the night). I'm still also still running the RX 570 system with a single thread, worksize of 8, and 768 intensity for now, mainly because I want to see how the other rig behaves before wasting more time fiddling with this one, but I suspect I will be doing the same tweak in a few hours.
It only made sense that things went the way they did (more power draw, harder on hardware, etc), because its the only way to overwhelm the FPGAs. ...
Totally agree - I have slowly but surely come around to the idea that it is the energy cost that determines how secure a network is to the first degree, not the absolute hashrate or cost of the mining hardware, per se. One of the coins I was an early supporter of that remained on the original CN algo (DERO) has a network hashrate in the range of 400-500MH/s which works out to around 2000 ASICs and 1.2MW total power draw. Prior to being ASIC'ed it had a network hashrate of around 4MH/s which works out to ~5300 RX 570 GPUs drawing ~800kW of power; hardly an improvement in network security, really, at the cost of alienating a massive number of previously loyal GPU and CPU miners, but I digress...
UPDATE: Hashrate dropped on one of the RX 560 (not the same one) yet again, then the display driver crashed so I decided to DDU 18.6.1 and install 18.5.1, then pixel patch and set all the cards to compute mode. I also stopped using OverDrivenTool and switched over to MSI AB, as v4.5.0 is once again able to handle more than 3 AMD GPUs. I have core set to 1200, mem set to 1900, power limit set to -20%, no change as of yet to core voltage, and total draw is at an unhealthy 375W with hashrate at 1740H/s. So a slight drop in hashrate, a rather higher increase in power draw, but if it's more stable I'll consider that a fair trade overall.
UPDATE 2: Well, I finally gave up on xmr-stak on the 4x RX 560 rig because it kept losing hashrate on one card then outright crashing the driver if I ignored that long enough. Even though I had no luck with SRBMiner 1.6.8 before, I decided to give it another go since I downgraded the driver to 18.5.1 and whaddyaknow, it now works. It's only been running about 10 minutes so far, but delivering a fairly consistent 1775H/s at 356W; so, a little better hashrate and a little lower power draw. Neither are enough of an improvement to justify the 0.85% dev fee, but if it can stay up for longer than 4 hours without needing a reboot then that will be worth the fee to me. I'd like to stick with xmr-stak on both my rigs, but v2.5.1 clearly needs some bug-fixes and/or additional tweaking.