Optimizing at Scale: Investigating Hidden Bottlenecks for Multi-Node Workloads

Accelerated computing (high performance computing plus high-throughput computing) promises to help us solve the toughest challenges of our time. However, scaling codes to run across multiple GPUs and server nodes is increasingly difficult. Varying hardware topologies, networking stacks, and storage backends add even more challenges.

We'll show the performance gains that can be achieved in these environments through NVIDIA Nsight Systems. Learn how Nsight Systems can help users identify bottlenecks, investigate their causes, and support developers as they modify their software. The amount of profiling data from a cluster of servers and network fabric can be immense. Nsight Systems' new multi-report analysis framework can help sift through data more rapidly via recipes for understanding performance and consistency. This will enable users to skip the guesswork and jump straight to the timeline reports of the most relevant ranks and time ranges.