Search papers, labs, and topics across Lattice.
This paper introduces campaign diagrams, a novel visualization technique that enables phase-level analysis of resource utilization and bottlenecks in workloads by integrating compute throughput, memory bandwidth, traffic volume, and latency into a single visual representation. Unlike traditional tools that either oversimplify or provide excessive detail, campaign diagrams effectively illustrate both ideal performance bounds and actual kernel performance, revealing optimization opportunities. The authors validate this approach through case studies on low-rank GEMM and Mamba, uncovering insights such as the surprising benefit of reducing operational intensity for improved performance and identifying fusion and pipelining opportunities across workload phases.
Campaign diagrams reveal that reducing operational intensity can paradoxically enhance performance, challenging conventional optimization strategies.
We present campaign diagrams, a visualization technique for phase-level analysis of resource utilization and bottlenecks in modern workloads. Existing tools have a trade-off: rooflines aggregate a workload into a single point and lose all notion of time, while profilers and traces expose fine-grained events but obscure what bounds performance. Instead, a campaign diagram depicts compute throughput and memory bandwidth utilization, compute and memory traffic volume, and latency in a single figure. Since they can be generated from analytical models, simulations, or profiling data, campaign diagrams capture both ideal bounds and a kernel's achieved performance. We demonstrate them on two case studies: a low-rank GEMM, where they reveal the counterintuitive result that reducing operational intensity can improve end-to-end performance, and Mamba, where they expose fusion and pipelining opportunities across phases. In both cases, our visualization technique reveals optimization opportunities that are difficult to identify with rooflines or profilers alone.
NVIDIA