With that configuration, I would look at the 390x or Fury-x (current highest single chip GFLOP performance.)
Since you have a 40 lane CPU, you could do 2x 390X, and dedicate one GPU to display, and not plug any cables into the 2nd and enable it in Vegas for compute only. Important not to plug monitors into 2nd card as is the temptation.
AMD has announced their next line up of Polaris GPUs that will support 4K60P and h265 encode and decode. So waiting for the price drop on 300 series might be a good bang for buck.
NVME – upgrade to an M2 slot NVME supported device, or get a PCIe slotted NMVE SSD device that uses the 40 lanes side of the CPU to operate projects from, or at least boot drive. NVME has significant reduction in latency and can handle multiple disk operations simultaneously. This will off load the majority of the disk IO from the DMI bridge. Use SATA ports for longer term project storage.
Make sure the board BIOS is kept up to date, and the DDR4 RAM is running at the board maximum stable frequency. Maxing your memory bandwidth is a low cost upgrade that can be determined with “winsat mem” from an admin command prompt when the system is idle. You want to be able to move the data as quickly as possible between Memory, CPU, and GPU. Make sure all your memory dimms are matched for speed and latency. Single high density DIMM per memory channel is better than filling the memory slots.