hugo-cwpearson/content/talk/20200225c3sr/index.md

title, event, event_url, location, address, summary, abstract, date, date_end, all_day, publishDate, authors, tags, featured, image, links, url_code, url_pdf, url_slides, url_video, slides, projects, math

title	event	event_url	location	address	summary	abstract	date	date_end	all_day	publishDate	authors	tags	featured	image	links	url_code	url_pdf	url_slides	url_video	slides	projects	math
Node-Aware Stencil Communication for Heterogeneous Supercomputers	C3SR Bi-weekly Technical Seminar		Coordinated Science Lab 216	street city region postcode country 1308 W Main St Urbana IL 61801 United States	Optimizing Multi-GPU Stencil Communication	High-performance distributed computing systems increasingly feature nodes that have multiple CPU sockets and multiple GPUs. The communication bandwidth between these components is non-uniform. Furthermore, these systems can expose different communication capabilities between these components. For communication-heavy applications, optimally using these capabilities is challenging and essential for performance. This work presents approaches for automatic data placement and communication implementation for 3D stencil codes on multi-GPU nodes with non-homogeneous communication performance and capabilities. Benchmarking results in the Summit system show that choices in placement can result in a 20% improvements in single-node exchange, and communication specialization can yield a further 6x improvement in exchange time in a single node, and a 16% improvement at 1536 GPUs.	2020-02-28T14:00:00Z	2020-02-28T15:00:00Z	false	2020-03-15T00:00:00Z	Carl Pearson		true	caption focal_point Image credit: [**Unsplash**](https://unsplash.com/photos/bzdhc5b3Bxs) Right				pdf/20200228_c3sr_stencil.pdf		example	internal-project	true