AM69A: Processing bandwidth, ECC calculations and etc..

Hi Ernest,

To give a little context before going through the questions, AM69A SDK builds on top of AM69 SDK, so there will be references to both SDK. For starters:

For 1 and 2, I would recommend looking through the performance guide in our SDK here: https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-am69/09_00_00_06/exports/docs/devices/J7_Family/linux/Release_Specific_Performance_Guide.html. In a different section of the same documentation, there is a network section here which briefly goes through DMA usage: https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-am69/09_00_00_06/exports/docs/linux/Foundational_Components/Kernel/Kernel_Drivers/Network/CPSW2g.html.

For 3, I would recommend the TDA4VM Academy page. Camera application is the main demos we have, but we also have a time-series data example: https://dev.ti.com/tirex/explore/node?node=A__ATxctZkqVXXEy4pDsNnupQ__TDA4VM-ACADEMY__dS93ClC__LATEST. The Academy goes through a lot more examples here: https://dev.ti.com/tirex/explore/node?node=A__AAkmgPumnb6pg20NWtipWA__com.ti.Jacinto%20EdgeAI%20Demos__PfvSod2__LATEST, and also has a lot of hands-on oriented how-to user guides, so I highly recommend going through these.

For 4, I am not quite sure if I understood the question correctly, but in our 8.6 SDK for AM69A we have an example pipeline using RTSP as an input: https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-edgeai/AM69A/08_06_01/exports/docs/common/edgeai_dataflows.html#rtsp-source. If customer already has an SK-AM69x board, then they can try this example out.

For 5, I do not know if there is a limit. Personally, I have tried out up to 512GB and have not seen an issue. From the SoC's perspective, it supports PCIe and USB, so as long as the driver for the connected device is available for Linux, then the device should be functional.

For 6, the model itself is saved on the micro SD card in our example, but should work for any other supported storage device for AM69A. The model is trained on a dataset on a host PC, so the micro SD card does not need to have the full dataset on the device as it only does inference. I would recommend taking a look here: https://github.com/TexasInstruments/edgeai-tidl-tools#introduction

For 7, we do not open up the R5F cores in AM69A. For details, I would recommend looking at the key differences here which also links to a build sheet that tell what is supported/unsupported: https://software-dl.ti.com/jacinto7/esd/processor-sdk-linux-edgeai/AM69A/09_00_00/exports/edgeai_docs/devices/AM69A/linux/faq.html. If wanting to use the R5F cores, then I would recommend using the TDA4x counterpart to the AM6x processors: https://www.ti.com/tool/J784S4XEVM

Lastly, for 8, I would recommend going through our SDK as the tools required are highly dependent on customer application. https://www.ti.com/tool/PROCESSOR-SDK-AM69A

Hopefully this answered most of the questions. Please go through these documentation I have linked and let us know if there is a need for clarification.

Regards,

Takuma