CoreSolver
CoreSolver is a private GPU-first solver we developed to compute postflop solutions for no-limit hold'em. Solving complex game trees generally relies on neural networks to reduce the required computational resources, at the expense of accuracy. Instead, we have opted for traditional solver algorithms to guarantee the greatest precision and reliability. Think of our solver as your favorite desktop solver, GTO+ or PioSOLVER, but with orders of magnitude more compute.
High-Precision Solves
Poker solutions are evaluated by their exploitability, a metric representing how much an ideal player could exploit the strategy. Exploitability is generally measured in terms of percentage of pot and is labeled dEV (delta-EV). Most established presolved libraries offer solutions in the realm of 0.2% to 0.5% dEV.
| dEV | bb/100 (5.5bb pot) | bb/100 (25bb pot) |
|---|---|---|
| 1% | 5.5bb/100 | 25bb/100 |
| 0.5% | 2.8bb/100 | 12.5bb/100 |
| 0.2% | 1.1bb/100 | 5bb/100 |
| 0.1% | 0.55bb/100 | 2.5bb/100 |
| 0.05% | 0.28bb/100 | 1.25bb/100 |
| 0.02% | 0.11bb/100 | 0.5bb/100 |
| 0.01% | 0.06bb/100 | 0.25bb/100 |
Common dEV% values and exploitability in bb/100
Our solver is built for GPUs, which have the raw power to efficiently compute solutions to dEV values as low as 0.01%. This level of precision is not necessarily required by everyone, but for serious students of the game and those building solution libraries, lower exploitability is strictly better than less converged solutions. For example, our testing has revealed that some of the solver's mixing of bet sizes completely disappears as solves reach ultra-low exploitability values. See our sample solutions to observe these differences yourself.
Commercial-Grade Libraries Under 24 Hours
We are currently targeting customers interested in custom solutions, from a few flops to full packages of all 1,755 strategically unique flops. We can scale horizontally across cloud GPUs to deliver full libraries of solutions in a matter of hours to days, rather than the weeks to months it normally takes to generate library-quality solutions at scale. For questions, inquiries, or raw sample data, email [email protected] or come join our Discord server.