Relying on cloud computing can lead to out-of-control costs. In order to solve this problem, researchers from Sky Computing Lab, a laboratory in Berkeley, California, launched an open source framework, SkyPilot, to help users shop around to reduce cloud computing costs. SkyPilot uses an algorithm to determine which cloud service region or provider is the best deal for a given project. The program takes into account the resource requirements of the workload (whether it requires a CPU, GPU, or TPU), and then automatically determines which cloud service region has the computing resources available to complete the work, sending it to the least expensive option for execution. SkyPilot Developer and Postdoc
For example, Zongheng Yang said that Azure currently offers the cheapest Nvidia A100 GPU instance, while Google Cloud and AWS offer premiums of 8% and 20%, respectively. The price difference for CPUs can even exceed 50%.
This article is transferred from: https://www.solidot.org/story?sid=73640
This site is only for collection, and the copyright belongs to the original author.