Profiling & Optimization

Learn how to analyze the performance of workloads running on Neuron devices and optimize compute with custom kernels.

---

• Neuron Explorer

  ---

  Visually analyze performance bottlenecks using Neuron Profiler and Explorer tools

  View Tutorial

• NKI Kernels

  ---

  Write custom high-performance kernels using the Neuron Kernel Interface (NKI)

  View Tutorial

---

Optimization Workflow

graph LR
    A[Run Workload] --> B[Collect Profile]
    B --> C[Analyze in Explorer]
    C --> D{Bottleneck Found?}
    D -->|Compute Bound| E[Write NKI Kernel]
    D -->|Memory Bound| F[Optimize Tensor Layout]
    D -->|Communication Bound| G[Adjust Distribution Strategy]
    E --> A
    F --> A
    G --> A