Hawkeye Optimizes GPU Kernels with Minimal Supervision
simonguozirui · x · 2026-08-22
Hawkeye is a hardware-aware GPU kernel optimization system. It allows experts to encode hardware tricks via unit tests, enabling AI agents to compose them into fast CUDA or HIP kernels for new architectures. The tool works across NVIDIA (Ampere, Hopper, Blackwell) and AMD (MI350) GPUs, often matching or beating expert-tuned baselines.
Related event: Hawkeye Enables AI Agents to Perform Hardware-Aware GPU Kernel Optimization(3 posts)→
More from Infra
- Optics Won’t Scale as Fast as Market Expects: Manufacturing Constraints — BenBajarin · 2026-08-22
- Nvidia uses linear algebra to fix cross-model KV cache transfer, 25x faster — bendee983 · 2026-08-22
- Anthropic hires former Google TPU head Amir Salek to boost in-house chip development — nmasc_ · 2026-08-22
- Lambda runs All-Reduce across 10,368 NVIDIA GB300 GPUs — TheZachMueller · 2026-08-22
- Running 284B DeepSeek on 64GB MacBook: Lossless quality confirmed — cowboy-bebob · 2026-08-22
- Dev Debate: Apple Silicon Remains Best for On-Device LLMs — walkingriver · 2026-08-22