MIT HAN Lab
Official@mit-han-lab · MIT
Efficient AI Computing. PI: Song Han
Agent Skills by MIT HAN Lab
Showing 2 vetted skills indexed across 2 GitHub repositories.
Frequently Asked Questions About MIT HAN Lab
FAQPage SchemaWhat specific performance tasks does MIT HAN Lab support?▼
The lab provides expert guidance for optimizing CUDA kernels specifically for NVIDIA Blackwell and Hopper architectures. It enables developers to profile kernel execution, identify performance bottlenecks, and implement hardware-specific optimizations to maximize throughput on high-end GPU hardware.
Which engineering personas benefit from these optimization skills?▼
These skills are designed for performance engineers, GPU architects, and systems researchers working on high-performance computing. Professionals focused on low-level hardware acceleration and those tasked with squeezing maximum efficiency out of NVIDIA B200 or Hopper-based clusters will find these resources essential.
What are the prerequisites for using these optimization reports?▼
Users must have access to NVIDIA Nsight Compute to generate the necessary profiling data. The process requires a functional CUDA environment and specific hardware targets, namely Blackwell B200 or Hopper architectures, to ensure the generated optimization reports are technically accurate and actionable.