KubeCon + CloudNativeCon + Open Source Summit + AI_dev China 2024

AI has penetrated into various industries, and companies have purchased many expensive AI GPU devices and used them for training and inference. So what is the reality of the use of these devices? Is the usage rate really high? Is the GPU card being monopolized by a large number of applications that are not heavily used? Do these AI devices work efficiently 24/7? This session will combine our mass production practices to summarize N ways to improve the utilization rate of AI equipment, such as * How to avoid monopoly and improve GPU usage through GPU sharing technology * How to improve GPU device usage through co-located in scenes with obvious tides * How to better perform GPU mark group matching training and inference applications to improve GPU usage This session will combine the practical experience of the two open source projects HAMi and Volcano in production, hoping to give everyone a clearer understanding of how to improve GPU usage.

人工智能已经渗透到各个行业，公司购买了许多昂贵的人工智能GPU设备，并将它们用于训练和推理。那么这些设备的使用情况如何呢？使用率真的很高吗？GPU卡是否被大量不常用的应用程序垄断？这些人工智能设备是否能够24/7高效工作？ 本场演讲将结合我们的大规模生产实践，总结提高人工智能设备利用率的N种方法，例如： * 如何通过GPU共享技术避免垄断并提高GPU使用率 * 如何通过与明显潮汐场景共同使用GPU设备来提高GPU使用率 * 如何更好地执行GPU标记组匹配训练和推理应用程序，以提高GPU使用率 本场演讲将结合两个开源项目HAMi和Volcano在生产中的实际经验，希望能让大家更清楚地了解如何提高GPU使用率。