CUDACyclone-SecretAdmirere: a Windows CUDA solver for the Bitcoin puzzle transaction
Hello everyone,
I rewrote a version of a CUDA puzzle solver, and I'm releasing it as open source (GPLv3).
Features- Compressed P2PKH addresses only
- Windows 10/11 x64 only, NVIDIA RTX 30xx, 40xx and 50xx GPUs only
- Randomized search (not a sequential scan)
- Modular design with easy-to-follow logic. The GPU math (CUDAMath.cuh) and hashing (CUDAHash.cuh) are separate headers. If you don't like my search logic, you are welcome to take these two files, or parts of the code, and use them in your own project under the GPLv3 license. The optimized math, hashing and kernels are where the speed increase on GPUs comes from.
SpeedI measured about
+27% over the reference figure from
btcpuzzle.info/benchmark on an RTX 3070 Ti (real kernel speed, not Speed With Included Overhead). I have only tested that one card so far. The other values in the README table are
estimates (the same 27% applied to the reference figure), and I will replace them as people post results. Please post your GPU, --grid and speed below.
RTX 5090 - 10,236 Mkeys/s Estimated
RTX 4090 - 7,569 Mkeys/s Estimated
RTX 3090 Ti - 3,581 Mkeys/s Estimated
Notes- It is a probabilistic search: there is no guaranteed coverage and no guaranteed time to find a key.
- --pubkey mode is intended for benchmarking only.
- The precompiled .exe may be flagged by some antivirus programs. The full source is in the repo, so you can build it yourself with Visual Studio 2022 and the latest CUDA Toolkit.
Download / sourceGitHub: CUDACyclone-SecretAdmirereCreditsInspired by Dookoo2's CUDACyclone. Includes parts derived from Jean Luc PONS' VanitySearch and RetiredCoder's RCKangaroo (both GPLv3).
DonationsMy PC was recently destroyed by a PSU failure and I'm saving for used parts to keep developing (about $2,000 needed). Any help is appreciated:
bc1q67nxvkge5ylq7fsvkhujmpvnut2a3964jqn4el
Use at your own risk.
Feedback, bug reports, suggestions, thoughts are welcome.