Optimizing Energy Efficiency for Applications on NVIDIA GPUs

Learn how to minimize energy consumption for any application running on NVIDIA GPUs. With increasing energy costs and environmental impact, it's increasingly important to tune for energy minimization, rather than solely focusing on the more traditional performance metric of time-to-solution. NVIDIA provides facilities to restrict GPU clock frequency and power draw: while this increases the time-to-solution, it allows overall energy usage to be tuned by finding the "sweet spot" for a specific workload. We'll demonstrate this technique for a range of applications on the latest Hopper GPUs, giving you the necessary information to tune your application's energy use without needing to change any source code. We'll discuss the trade-offs between tuning for energy versus time-to-solution, and also provide comparisons across applications to give guidance toward settings suitable for overall energy efficiency for mixed workloads.