Benchmarks

Powerful, power-efficient performance for edge AI

Perceive Ergo® AI processors enable product developers to leverage large neural networks capable of delivering advanced features, while operating within tight power constraints. The following table provides some performance and power-efficiency benchmarks, based on our own testing with several well-known neural networks.
ErgoErgo 2
Max App IPS*Compute Power @ 30 fps (mW)Power Efficiency (IPS/W)Max App IPSCompute Power @ 30 fps (mW)Power Efficiency (IPS/W)
Image Classification
MobileNet-v2280932841106192631
ResNet-50262112727979172465
Object Detection
YOLO-v5-S445060011575458

* Inferences per second

Don't see your network here?

This is not an exhaustive list of the neural networks supported by the Ergo product line – please get in touch with us to discuss your specific needs.

Notes:

  • On nominal chips, at nominal voltage, at room temperature, Dec. 20, 2022.

  • MNv2 and RN-50 on ImageNet; YOLO-v5 on CoCo

  • Ergo at 250 MHz, Ergo 2 at 333 MHz.

  • Margin of error +/-10%.