GPU Optimization Patterns

Expert patterns for memory management, torch.compile, profiling, quantization, and kernel optimization.

Pattern 1: GPU Memory Profiling and Budgeting

Profile and track GPU memory usage with explicit memory math.

import torch

def memory_report(label: str = ""):
    allocated = torch.cuda.memory_allocated() / 1e9
    reserved = torch.cuda.memory_reserved() / 1e9
    print(f"[{label}] Allocated: {allocated:.2f} GB | Reserved: {reserved:.2f} GB")