PCIe Mode App

In PCIe mode, once a pipeline is developed, you must create a host-side application and package the MPK for deployment on the host system. SiMa provides C++ and Python APIs on Ubuntu x86 platforms to streamline host application development.

Starting with SDK 1.7, you can also use the Co-Processing API to deploy and manage pipeline lifecycles, as well as perform device firmware updates programmatically.

Note

These APIs become accessible after installing the PCIe driver.

In the PCIe mode, you can currently use the Machine Learning Accelerator (MLA) to run inference tasks (quant -> NN Model -> dequant). Additionally, with our support, you can manually generate the MPK in the SDK to enable any valid GStreamer PCIe pipeline.

In a future release, this mode will expand to include access to all hardware blocks, such as video codecs, enabling pre- and post-processing operations directly on the MLSoC. This enhancement is part of our ongoing roadmap.

Refer to this example to create a host-side application that leverages the SiMa MLSoC for running a ResNet-50 workload.