Tag
This article argues that modern hardware, particularly IOMMUs, makes microkernels viable again by eliminating overhead from context switches, and suggests revisiting microkernel designs for better security and reliability.
KDA (Kernel Design Agents) is an open-source agent-centric workflow from MIT HAN Lab for using coding agents to research, implement, verify, and iterate on performance-sensitive CUDA kernel tasks. It helped Databricks achieve rank #1 on NVIDIA's SOL-ExecBench leaderboard.
Enze Xie announces Sol Video Inference Engine, an agent-native, training-free full-stack accelerator for video diffusion that auto-tunes cache, sparse attention, token pruning, quantization, and kernel fusion, achieving >2× end-to-end speedup on large models like 64B Cosmos3-Super and 22B LTX-2.3.
Implemented a top-k kernel from the kernel design section of the MSA paper, using exp-free comparison and warp-level tree merging with CUDA shuffles. The code is available on GitHub.
KDA is an agent-driven kernel design framework that helped HAN Lab achieve top rankings in the MLSys FlashInfer Kernel Contest by minimizing human involvement. The agent leverages Humanize, KernelWiki, and profiler skills to produce state-of-the-art kernels.
This opinion piece argues that AI kernel portability across different hardware (TPU, GPU, etc.) is structurally impossible due to fundamental hardware differences, and that the best AI stacks will always require hardware-specific DSLs for optimal performance, despite the industry's desire for portability.