Your Dice Collection
Click + to add dice you own. Need at least 6 to see loadout rankings.
Expected score is the average points per turn for that loadout (from 100k simulated turns using optimal banking).
Farkle% is how often a turn ended with no bankable dice — lower is safer, but some high-scoring loadouts farkle more often.
Top loadouts across all 2.3 million possible combinations, ranked by expected score per turn.
How Scoring Works
Devil's Head Die is a joker — it takes the face value that maximises your score each roll. The simulation uses a greedy strategy with dice-remaining-aware banking thresholds, optimized per loadout.
About This Tool
This concept started as a Python-based Monte Carlo simulation. Over time it became clear we could just compute all possible combinations upfront and avoid having users run the simulation on their machine. Every possible 6-die loadout (2,324,784 combinations from 32 distinct die profiles) is now simulated once using a CUDA-accelerated Monte Carlo engine. Each loadout was run 100,000 times; results and farkle rates are pre-computed — filtering is instant and runs entirely in your browser.
Banking Strategy: Dice-Remaining-Aware Thresholds
When you keep rolling with fewer dice left, farkle risk rises sharply. A single fixed “bank at 350” rule fails for both weak and strong loadouts. We use dice-remaining-aware thresholds: the bank point scales with how many dice you have left. With 6 dice you might need 500 to bank; with 1 die left, the threshold drops to about 3% of that. This matches real risk.
Optimal Strategy Per Loadout
We ran a sweep across 12 different base thresholds (200 to 8000) for every loadout. Weak dice like 6× Ordinary Die perform best with conservative banking (bank early, around 250–300). Strong dice like 6× Weighted Die perform best with aggressive banking (push for 5000+). Each loadout uses its own optimal threshold in the final scores.
Farkle Rates & Risk
Some loadouts score high but farkle often. We record the fraction of turns that result in a farkle so you can weigh reward vs risk. A safer set with a slightly lower expected score may be preferable if it has much lower farkle chance.
Scoring Update (Mar 2026)
We cross-checked against KCD2 guides (IGN, kcd2.org) and added two previously missing 5-dice straights: 1–5 (500 pts) and 2–6 (750 pts). The full simulation was re-run with the corrected scoring; all 2.3M loadouts now reflect accurate KCD2 rules.
AI-Assisted Development
CUDA optimization: AI was used to optimize the simulation engine, improving GPU utilization and cutting run times significantly. That speedup allowed many more iterations per loadout, which in turn improved the search for optimal banking thresholds and more reliable expected scores.
Site UX: AI also helped refine the web interface — the dice grid, results layout, upgrade suggestions, and overall flow — to make the tool smoother and easier to use.