Optimize Deep learning GPUs performance through iterative profiling

NVIDIA Nsight systems is a performance analysis tool, when applying to deep learning applications, it is important to first understand the application’s workflow, in the context of deep learning training job launch, in order to iteratively optimize via profiling techniques such as nvtx tags complimenting NVIDIA Nsight systems.

In this talk, we are going to leverage NVIDIA’s profiling toolchains anchored in a deep learning training job launch scenario and go through one iteration for optimization and visualize generic stages of deep learning training job launch, namely GPU(s) initializing, data loading, and model training and checkpointing.
The concept is generically applicable across a variety of deep learning training optimization for single and/or multiple GPUs.

Bio:
Zenodia Charpy is a Senior Data Scientist and solution architect working at NVIDIA, she is working partially within NVIDIA Applied NLP Research team collaborating with hundreds of researchers around the globe.
She has 8 years of solid experience as a data scientist solving real-world problems, and building end-to-end solutions both as an in-house data scientist as well as a data science consultant.

Zenodia Charpy

Her kan du læse mere om Zenodia Charpy, som holder et oplæg med fokus på Deep learning. Mød Zenodia Charpy til dette års Driving IT.

Link kopieret

Zenodia Charpy

Driving IT

Tilmeld dig konferencen