English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Scientists are increasingly exploring and utilizing the massive parallelismof general-purpose accelerators such as GPUs for scientific breakthroughs. As aresult, datacenters, hyperscalers, national computing centers, andsupercomputers have procured hardware to support this evolving applicationparadigm. These systems contain hundreds to tens of thousands of accelerators,enabling peta- and exa-scale levels of compute for scientific workloads. Recentwork demonstrated that power management (PM) can impact application performancein CPU-based HPC systems, even when machines have the same architecture and SKU(stock keeping unit). This variation occurs due to manufacturing variabilityand the chip's PM. However, while modern HPC systems widely employ acceleratorssuch as GPUs, it is unclear how much this variability affects applications.Accordingly, we seek to characterize the extent of variation due to GPU PM inmodern HPC and supercomputing systems. We study a variety of applications thatstress different GPU components on five large-scale computing centers withmodern GPUs: Oak Ridge's Summit, Sandia's Vortex, TACC's Frontera and Longhorn,and Livermore's Corona. These clusters use a variety of cooling methods and GPUvendors. In total, we collect over 18,800 hours of data across more than 90% ofthe GPUs in these clusters. Regardless of the application, cluster, GPU vendor,and cooling method, our results show significant variation: 8% (max 22%)average performance variation even though the GPU architecture and vendor SKUare identical within each cluster, with outliers up to 1.5X slower than themedian GPU. These results highlight the difficulty in efficiently usingexisting GPU clusters for modern HPC and scientific workloads, and the need toembrace variability in future accelerator-based systems.

Not All GPUs Are Created Equal: Characterizing Variability in Large-Scale, Accelerator-Rich Systems