Show HN: CUDA Profiler for Production Inference

Category: observability

Tags: cuda-profiler, inference-optimization, llm-tracing

Score: 6.8/10 (Innovation: 6, Technical: 7, Documentation: 7, Utility: 7)

Graphsignal is a production-scale inference profiler for AI models, providing high-resolution timelines, LLM tracing, and system metrics across GPUs and accelerators. Its integration with AI coding agents for optimization and low-overhead CUPTI-based profiling makes it interesting for performance engineering in AI deployments.

Target audience: ai engineers, ml engineers, devops

Repository: https://github.com/graphsignal/graphsignal-profiler · Python · Apache-2.0 · 207 stars

View on Hacker News