Reconfigurability of Jetson AGX Xavier

[Halima_Bouzidi]

Hello,

I’m currently working on Jetson AGX Xavier to run ML-based applications. Specifically, I’m studying the reconfigurability of the Jetson AGX Xavier to deliver better performances. So far, the number of processing units, operating frequencies of CPU, GPU, EMC, and DLA are adjustable according to the “nvpmodel.conf” file. However, I’m wondering if it’s also possible to adjust other parameters of the NVDLA accelerator? For instance, is it possible to tune some parameters of the convolutional buffer, core, and the memory settings? Moreover, what are the possible frequency values of the DLA that can be written in the “nvpmodel.conf” file?

I really appreciate any help you can provide.

[AastaLLL]

Hi,

nvpmodel and jetson_clock doesn’t set DLA hardware parameter.

Currently, please use TensorRT API to deploy a job to DLA.
We don’t provide a tool to set the hardware parameter of DLA.

Thanks.