Before commenting on performance, the stuttering that was present in #660 and black frame issues in the the older version of #667 are no longer present.

Testing with 1440p host resolution, capture 1440p/60 @ 90MBps. GravityMark's real performance stays in the range of 80-130fps, average ~110fps (never any major dips).

• Vulkan: average capture is ~55fps with two very noticeable dips down to 35fps (asteroid field close-ups).
• D3D12: average capture is 45-50fps.
• D3D11: average capture is solid 60fps (very small dips on scene change)

Further testing with AMD's built-in Chill framerate limiter:

• Vulkan via Chill @ 60fps min/max: capture rate runs at a solid 60fps with no noticeable dips.
• Vulkan via Chill @ 80fps min/max: capture rate runs at a solid 60fps, perhaps or or two minor dips to 55fps.
• D3D11 via Chill @ 60fps min/max: capture rate runs at ~50fps with frequent dips [N.B. this also happens on latest nightly, so it's not an indication of something wrong with this PR]
• D3D11 via Chill @ 80fps min/max: capture rate runs at solid 60fps.
• D3D12: n/a (Chill cannot limit framerate in GravityMark with this API)

In summary, it seems 1) uncapped rates may be causing GPU contention that results in the encoder throttling in Vulkan and D3D12 (but not D3D11), and 2) the D3D11 capped @ 60fps tests result result seems quite counter-intuitive (but the poor performance in that case is unrelated to this PR).

I need to test some more with other non-Vulkan applications, but so far it looks very promising and is definitely an improvement compared to nightly.

Updated PR (with call to SetGPUThreadPriority() for second device) seems to perform more or less the same as before.

I also tested some 4K/60 host+client capture. Current nightly will slightly miss the target, hitting about 57fps in Cyberpunk 2077 or even when browsing https://www.vsynctester.com/, but the PR improves the capture rate to reach the 60fps target. Will continue testing and let you know if I encounter any issues.

Edit: I also tried pushing GravityMark a little further. At 4K on both client and host, Vulkan capture runs at mostly 60fps with just a few dips to 30 fps if the host is capped to 60FPS. The worst point is the end of the demo; the host drops to ~55fps and the capture rate is halved to ~27fps. If uncapped, host performance ranges from 55-70fps, and capture performance is worse than 1440P (~25fps average). So for AMD, GPU load definitely influences capture performance.

This PR is ready for merge.