Blackwell NVFP4 Kernel Hackathon

​Overview

​Join the Developer Kernel Hackathon, a four-part performance challenge hosted by NVIDIA in collaboration with GPU MODE.

​This event invites developers to push the limits of GPU performance and optimize low-level kernels for maximum efficiency on NVIDIA Blackwell hardware.

​Across four problems released throughout the hackathon, participants will compete to author kernels that approach the speed of light.

​Whether you’re a seasoned kernel developer or just eager to test your limits, this hackathon offers the chance to showcase your expertise and join a community of world-class developers.

​Special thanks to our partners:
Sesterce, a high-performance GPU cloud platform, is contributing DGX B200 compute resources to support participants throughout the competition.

​Dell is providing a Dell Pro Max with GB300 as the grand prize.

​Schedule

​Each kernel problem will be released sequentially. Once one problem ends, another one begins.

• ​Kernel #1 - NVFP4 Batched GEMV

  • ​Nov 10th - Nov 28th

• ​Kernel #2 - NVFP4 GEMM

  • ​Nov 29th - Dec 19th

• ​Kernel #3 - NVFP4 Gated Dual GEMM

  • ​Dec 20th - Jan 16th

• ​Kernel #4 - NVFP4 Grouped GEMM

  • ​Jan 17th - Feb 13th

​How to Participate

• ​Open to individuals only (no teams).

• ​Register through this page by February 13th, 2026 to be eligible to win prizes.

• ​Join the GPU MODE Discord, and head to the nvidia-competition channel for announcements, discussions, questions, and assistance.

• ​Submissions can be made using the Popcorn CLI. Follow the setup and submission instructions here:
  👉 https://github.com/gpu-mode/popcorn-cli

​Prizes

​Each of the four kernel problems will have 3 winners who will receive latest generation NVIDIA hardware, with one grand prize winner across all problems for achieving performance closest to the speed of light.

​The top 2 winners of each problem will also be invited to a special awards ceremony at GTC in San Jose, March 2026.

​💥 Grand Prize:
1× Dell Pro Max with GB300 + GTC 2026 Pass – awarded to the participant whose submission (across any of the four problems) achieves performance closest to the speed of light.

​🏆 Prizes for Each Kernel Problem:
There are four kernel problems in total, and each will have its own set of winners:

​🥇 1st Place: NVIDIA DGX Spark + GTC 2026 Pass
🥈 2nd Place: NVIDIA RTX 5090 + GTC 2026 Pass
🥉 3rd Place: NVIDIA RTX 5080

​Scoring and Judging

• ​There will be four independent problems, each scored separately.

• ​The top 3 submissions for each problem will win prizes.

• ​The grand prize will be awarded to the participant with the fastest kernel overall, measured by closeness to the published “speed of light” performance for that specific kernel problem.

• ​Submissions are benchmarked on NVIDIA GPUs using the GPU MODE infrastructure.

• ​Scoring is based on absolute runtime and relative speed-up against baseline reference kernels.

​Additional Resources

​For learning resources, check out and subscribe to GPU MODE YouTube channel, where you can find weekly lectures from top voices in the ML community, including researchers and engineers from NVIDIA.

​Terms & Conditions

​Participation in this hackathon is subject to the official terms and conditions.
🔗 View full Terms & Conditions