The miner progressively shows/detects less and less memory for detected GPUs
When the miner starts it shows 8148 MB for each card, and it goes all the way down to 1000 MB per card. This can happen in an hour or in 12 hours. Stopping and restarting the miner fixes the problem.

OS: Ubuntu 16.04.1
AMD drivers: 16.30
All cards are ASUS ROG STRIX-RX480-8G-GAMING

Has anyone else experienced this?
@Claymore: Could it be that you are not releasing GPU memory between internal miner restarts?