深入理解 Nsight System 与 Nsight Compute 性能分析优化工具

BingLiu, 202012DEEP DIVE INTO NSIGHTSYSTEMS & NSIGHTCOMPUTE2AGENDAOverview of ProfilersNsightSystemsNsightComputeCase StudiesSummary3OVERVIEW OF PROFILERSNVVP Visual Profilernvprofthe command-line profilerNsightSystems A system-wide performance analysis toolNsightCompute An interactive kernel profiler for CUDA applicationsNote thatVisual Profilerandnvprofwill be deprecatedin a future CUDA releaseWe strongly recommend you transfer to NsightSystems and NsightCompute4NSIGHTPRODUCT FAMILY5OVERVIEW OF OPTIMIZATION WORKFLOWInspect &AnalyzeOptimizeProfile ApplicationIterative process continues until desired performance is achieved6NSIGHTSYSTEMSOverviewSystem-wide application algorithm tuning•Focus on the application’s algorithm –a unique perspectiveLocate optimization opportunities•See gaps of unused CPU and GPU timeBalance your workload across multiple CPUs and GPUs•CPU algorithms, utilization, and thread state•GPU streams, kernels, memory transfers, etcSupport for Linux & Windows, x86-64 &