Quote from: ares-austria on May 22, 2023, 04:00:23 PM

In a quick comparison we can see that the "Dynex Neuromorphic Chip" SAT solver is actually dog slow compared to a single threaded YalSAT instance running on low-clocked server hardware: Dynex SAT solver vs. YalSAT benchmark.

Here YalSAT yielded close to 3 x performance compared to the Dynex sample. YMMV, so if you're into this kind of stuff - just do your own runs with the software:

DynexSolve (Dynex Neuromorphic Chip)
https://github.com/dynexcoin/Dynex-Neuromorphic-Chip

YalSAT
https://github.com/arminbiere/yalsat

Looks like the Dynex team has been noticing our criticism and have updated their repository with benchmarks that look to be in their favour. Unfortunately it is easy to see that they are comparing different things. The YalSAT is a single threaded Sequential Standalone Version, where as the Dynex SAT solver uses 8 threads as shown in their source code. This results in, once again, amazing results with their data.

However, there is a PalSAT (Parallel Simple Portfolio Version) version included in the code repository. With this one the results are once again much in favor of the Local Search solver over the Dynex Solver, with PalSAT clocking in results for their dataset at over 2.6 x faster results on average. We have opened a ticket on the project for the team to clarify and include PalSAT results to their index page: Dynex SAT benchmarks for YalSAT 1.0.1 include no multithreaded results

We look forward to seeing the updated results or at least a response from the team to the public ticket.