Why would you not go the other way with the fans now that it is in cased in a room?  Suck the air through the miners and exhaust them to the outside?

I have the same issue, gets stuck after switching to dev pool, unusable in current state... stops every couple hours.

That particular card that is having the drop out is 72.4% asic quality.  The other MSI armor 570 I got in that batch is 66% asic quality and is performing fine.  I am also not pushing the undervolting much, 850mv core & 900mv mem.

Running 2.8c now, One of my GPU's still drops to 0.0mh.  I have two models of this card, same memory, both with stable overclock.  Runs fine with 2.7b.  Have tried to lower significantly its memory clock from 2200 to 2100, 2075 and 2050, does not fix the problem, left the core at 1150.  With version 2.8c I have tried to use the -clnew 0 flag for that particular GPU while designating the other cards in the system -clnew 1 yet it does not fix the problem.  I run the card with -mi 12 on 2.7b so I also tried to lower that to -mi 10 on 2.8c with no difference in result.  Only way that card remains hashing is by running it separately with 2.7b.  I have also tried claymore 11.6 version with the same conclusion, hashes for a minute or two and drops out to 0.0mh. 

Has anyone else experienced this particular problem?  Its not a major issue for me running two versions but it is the idea that I cannot solve this problem has me twisted atm.   

Running 2.8b, One of my GPU's after finding a share its hashrate drops to 0.0mh, it starts at >30mh then as soon as a share is submitted with that particular card it drops out.  I have two models of this card, same memory, both with stable overclock.  One runs fine with 2.8b, the other has the issue.  I have tried changing my -mi for that specific card lower and the problem persists.  Running previous PM versions the card runs fine.  Is there a way within the same prompt to individually designate which kernal a specific card runs on?  If so is there an example config for my bat?