Awesome news, but on the other hand a bit underwhelming 
I’ll explain why.
It seems this Optix implementation works in a completely isolated way, totally separated from the rest of Cycles, so for example there is no CPU+GPU rendering, so in one powerful workstation (a 2990WX) if I need to render a scene such a machine I have to decide what will be faster, CPU+GPU without Optix or just the GPU (an RTX 2080 over CUDA has more or less the same speed as a 2990WX, so in the end 2990WX + RTX2080 could be faster in Master than the Optix version alone).
On the other hand, we have several optimizations that I’m not sure if they are compatible, like Scramble Distance, Dithered Sobol and Adaptive Sampling, in this case I’m not sure if they are compatible or not, I hope so, but if they are not, we could be loosing some super powerful optimizations that we already have that make Cycles waaaaay faster than the Cycles in 2.80 master as of today.
Finally there is the RAM problem, NVLink is not supported, for example it’s impossible to render a big project with Optix, our current project requires 20Gb of ram, and if we work in an exterior visualization that is not strange, I doubt that Optix could be used for something like Spring (I may be wrong, I’m not sure bout the total weight in RAM), but as of today, we are limited to 8Gb or 11Gb of ram (unless we have a Titan RTX with 24Gb) but it does not make use of the “sum up ram” capability of the bigger RTX GPU’s to sum up ram to get 16Gb, 22Gb or 48Gb. (Of course this will probably be solved in the future)
Don’t take me wrong, I know this is the first implementation and they are BIG GREAT NEWS, I just wanted to tell the things that worry me the most regarding the current Optix implementation, I hope in the future Optix will work under the hood, being compatible with all the other optimizations and with CPU + GPU + NVlink + Optix optimizations for GTX cards