Skip to content

GPU solving: the wall is memory, not arithmetic

Eternity II looks like the perfect GPU workload: millions of independent subtrees, embarrassingly parallel. Eighteen years of community attempts measured a different reality: divergent branching, per-thread state that outgrows fast memory, and a placement loop that was already memory-bound on CPUs. What the GPUs actually delivered, and the one workload where they genuinely shine.

conceptGo fasterGPU & hardwareUpdated 2026-07-02

Keep exploring

Page sourceView as Markdown