Profiling AI Workloads, and Beyond, with ROCm Systems Profiler

In the race to build and serve AI, every hardware optimization counts, and you can only optimize what you can see. This talk covers the core tracing capabilities of ROCm Systems Profiler, which brings CPU, GPU, network and runtime activity into a single system-wide view. It then shows how those capabilities apply to a real-world LLM workload: Llama inference on ROCm.
Format: ML APPLICATION
Location: ROOM B
Speakers: Aleksandar Djordjevic
Level: All levels
Audience: Registered participants