Getting Efficient DNN Inference Performance: Is It Really About the TOPS?

This presentation looks at how performance is measured among deep learning inference platforms, starting with the simple peak TOPS metric, why it’s used and why it might be misleading. Brown looks at compute efficiency as measured by real benchmark workload performance and how it relates to peak TOPS, comparing performance across Intel’s inference platforms. He also discusses how developers can use Intel’s DevCloud for the Edge to quickly access Intel’s inference platforms.

See here for a PDF of the slides.

To view the rest of the 2020 Embedded Vision Summit videos, visit the event's video archive.

Register for the 2021 Embedded Vision Summit here.