https://niyunsheng.github.io/series/performance-modeling/Recent content in Performance Modeling on Yunsheng NiHugo -- 0.154.5en-usContent is licensed under CC BY-NC-SA 4.0.Mon, 09 Feb 2026 23:20:00 +0800https://niyunsheng.github.io/roofline-llm-h200/Mon, 09 Feb 2026 23:20:00 +0800https://niyunsheng.github.io/roofline-llm-h200/A quantitative Roofline analysis of LLMs on NVIDIA H200. We derive compute-bound thresholds, analyze the 1:10 communication bottleneck, and propose optimal strategies for activation recomputation and operator fusion to maximize hardware efficiency.https://niyunsheng.github.io/llm-flops-mfu-hfu/Sat, 31 Jan 2026 09:30:00 +0800https://niyunsheng.github.io/llm-flops-mfu-hfu/A rigorous breakdown of FLOPs in Llama-style architectures, deriving the relationship between linear projections and quadratic attention overhead, with insights into sample packing efficiency.https://niyunsheng.github.io/gpu-network-constants/Sun, 25 Jan 2026 23:21:35 +0800https://niyunsheng.github.io/gpu-network-constants/A quick reference of Dense FLOPS and Unidirectional Bandwidth for A100, H100, H200, and Blackwell.