LeetGPU
Write fast GPU kernels in CUDA, Triton or PyTorch and rank on a leaderboard
LeetGPU is a GPU programming challenge site. Each problem, such as vector addition, matrix multiplication or attention, gives you a starter file and asks for a fast kernel. You can write it in CUDA, Triton, PyTorch, Mojo, CuTe DSL or JAX. Submissions are checked for correctness, then timed, and ranked on a leaderboard.
Freemium · Login needed · Browser · Intermediate
About LeetGPU
It trains parallel thinking and kernel tuning: how threads split the work, how memory is read and what each choice costs. The public challenge repo holds about 144 problems split into easy, medium and hard, so it suits newcomers to GPUs as well as people who tune kernels for work.
It runs in the browser, with NVIDIA GPUs from the T4 to the B200 to run on and a playground that simulates a GPU. You sign in to submit. Basic use is free, and a paid Pro plan adds features such as kernel profiling and AI help.
Made this game? Claim it →
Tips and discussion
No tips yet.
Sign in to share a tip.