KernelSprint: GPU Optimization by Bounty

Field notes

KernelSprint organizes GPU optimization as a contest-like procurement workflow: a team publishes the workload, baseline, target, reward, and deadline, while engineers submit competing kernels against the same brief. The page names CUDA, Triton, and PyTorch and says teams receive source code plus benchmark evidence after checking correctness, performance, and integration. Its first proposed challenge carries a $2,000 prize, but the site explicitly says that prize is not yet funded and that no bounties are currently posted.