It is a rig of 8 gtx 1060 6gb and yes B stand for billion. I have 2 of that type and right now I am building one with 2080 super.
Since we are now in the billions, we should also use x Bk/s, because 4000 Mk/s is shit. I better write 4 Bk/s. With my new rig I can expect about 10 Bk/s.
However I have significant speed loss if I am searching for a large number of prefixes. With about 10k prefixes I am down to 700 Mk/s, but I just parsed some data and want to search for 10M prefixes. I dont know where the speed will be with so many prefixes. In oclvanitygen there were no impact on the speed - doesnt matter if you are looking for 10 or 10M prefixes. But vanitysearch is still 5-6 times faster with so many prefixes.

I was thinking of doing an FPGA, but with this much speed I dont need an FPGA. I probably just put more rigs on this task.