FYI, tried to use the "boost" function on my 6x Vega56 @ 64 rig (for x16r) but it crashed almost instantaneously. 
Not going to try to "optimise" it for now either. It hashes fairly well (albeit at clocks that are lower than they should be, imo) so I'm not jeopardising the rig at the moment (until I'm certain the functionality works - afaik it's not officially supported on these Vega cards, especially on driver 18.6.1).
Not going to try to "optimise" it for now either. It hashes fairly well (albeit at clocks that are lower than they should be, imo) so I'm not jeopardising the rig at the moment (until I'm certain the functionality works - afaik it's not officially supported on these Vega cards, especially on driver 18.6.1).
I know, I know... The plan with the C++ version is to outsource testing and optimizing default settings.
My sponsor set up a pretty interesting arrangement for the project. We will see.
x16r is an algo I am sure you can still optimise greatly. The nVidia miners have gained about 50% speed over the last months, purely through software optimisation. [subtle hint]Currently my 6x Vega rig hashes slower than my 5x GTX1080 rig.[/subtle hint]
I haven't done any optimizations for X16R, so I don't think it is difficult to catch up with CUDA miners.
I have to take care of other algorithms first, though: CryptoNightHeavy, Lyra2REv2, NeoScrypt, and CryptoNightV7.