Put a bounty on your GPU bottleneck.
Post your workload, set a reward, and invite kernel engineers to make it faster. Receive source code you can benchmark in your own stack.
Agree on correctness, performance, and delivery before work starts. Engineers know what earns the reward. Your team knows what to test.
Write fast kernels? Browse open bounties.
How KernelSprint works
- 01Post a bountySet the workload, baseline, reward, and deadline.
- 02Submit kernelsEngineers develop candidates against the same brief.
- 03Review resultsCheck correctness, performance, and integration.
- 04Select & hand offApply the award terms. Deliver source and evidence.
One brief. Parallel approaches. A result you can reproduce. Evaluation and payouts are arranged manually.
CUDATritonPyTorch
For kernel engineers
Find your next optimization.
No bounties posted yet
Have a workload worth speeding up? Post the GPU, baseline, target, and reward so engineers can see where they can help.