KernelBot: The World's First Competitive GPU Programming Platform
summary
KernelBot runs a time‑shared GPU service with a CLI and Discord integration that yields sub-15s feedback (modal hits sub‑10s) and uses PyTorch tensors as the canonical input so everyone competes on the same correctness baseline. It collects ~60,000 kernels from ~600 users, holds multiple $100K competitions with AMD and Nvidia, and reports community submissions beating optimized commercial baselines; compile optimizations use NVRTC: 0.1s compiles, scheduling relies on GitHub Actions scheduler, and submission tooling is popcorn CLI. Security and eval‑gaming require active patching.
AI-written summary · about
chapters (10)
- 0:00KernelBot and GPU programming
- 1:27Motivation for competitive GPU
- 2:32Fast GPU workflows
- 3:54Platform features and UX
- 4:54Problem design and security
- 6:19GitHub Actions and profiling
- 7:20Compile times and NVRTC
- 8:23Company partnerships
- 9:46Community growth
- 10:27How to get started with KernelBot
related talks
18mKernelAgent: Hardware-Guided GPU Kernel Optimization Via Multi-Agen...
08mGPU Programming with Triton Kernels
19mAI Kernel Generation: What's working, what's not, what's next
25mLeveraging AI Ecosystem for Portable and Sustainable GPU Kernels in HPC
22mBridging the Hardware Gap With Code Harnesses on the Hugging Face Kernels Hub
29mBeyond CUDA: GPU Accelerated Computing on Cross-Vendor Graphics Cards with Vulk...
27mVendor-Neutral GPU Programming in Chapel
10mMaximizing Kernel Development Productivity Under Performance Constraints