nice! clocking ~18Sols/s on the same R9 390X that does ~22 using my own brew. your miner has one major advantage though...it actually comes up with solutions that are accepted by the pool 
An optimal solution should be able to do near 100 H/s (i.e. 200 sol/s) on a R9 380. With 2 million records and 32-64 bytes of memory IO per record * 10 rounds = 1.28GB of memory bandwidth per hash. The R9 380 with a memory clock of 1.5Ghz has 192GB/s of bandwidth. The hard part is making full use of that bandwidth when you have small random IO. Ethash is just at the fringe of being able to maximize memory bandwidth with 128-byte random reads. Since an equihash solver requires a lot of small reads and writes, the algorithm and data structures need to be optimized for the GPU memory architecture. i.e. 8 x 32-bit channels working with 2K pages and a 256-byte stride per wavefront.
I expect by the 28th we'll see GPU miners doing 50-60 sol/s, and close to 100 sol/s a month from now. Hopefully one of the faster ones will be mine. :-)
So its deff H/s not Sol/S to go off?
Everyone was telling me its the same thing or to go off Sol/s?
Is it better to wait off on these cards to see if some updates work better on other cards first?
Wagner's algorithm produces 2 solutions to the generalized birthday problem. Each equihash run produces 2 solutions, so 1 equihash/s = 2 solutions/s.
If you want to know the details, read the paper. It's a heavy read though. It wasn't until the 2nd or 3rd time I read it that I started to really understand the algorithm enough to start thinking about coding.
https://www.internetsociety.org/sites/default/files/blogs-media/equihash-asymmetric-proof-of-work-based-generalized-birthday-problem.pdf
Thank you
