This file has been truncated. show original
Progress report on the Cuckoo Cycle GPU solver
Since completing my rewrite of xenoncat's performance quadrupling CPU solver (winning a double bounty)
in the form of mean_miner.cpp, I've been slowly grinding away at porting that code to CUDA.
I consider myself an amateur GPU coder, having previously ported the latency bound lean_miner to CUDA
(and having to pay a bounty to fellow Dutchman Genoil for improving performance by merely
tweaking the threads-per-block which I had naively fixed at 1), as well as my own
[Equihash miner](https://github.com/tromp/equihash) submission to the
[Zcash Open Source Miner Challenge](https://z.cash/blog/open-source-miner-winners.html).
That Equihash CUDA miner achieved a paltry 27.2 Sol/s on an NVIDIA GTX 980,
matching the performance of my Equihash CPU solver. But it did serve as the basis for far more capable
rewrites such as this
by Leet Softdev (a.k.a. djezo), which achieves around 400 Sol/s on similar hardware.
Today, on Jan 30, 2018, I completed work on my CUDA solver, mean_miner.cu