How to use Deep Learning Accelerator

Autonomous Vehicles DRIVE AGX Xavier DRIVE AGX Xavier General
1

Hi,

There is such a statement “In this release of DRIVE Software, TensorRT does not include full Deep Learning
Accelerator (DLA) support and therefore it is not at full performance” in NVIDIA DRIVE Software 9.0. So, I’m confused about it, and How can I use it correctly?

Thanks

2

Dear denglong,[url][/url]
DLA can not be programmed unlike a GPU. You need to use TensorRT to make use of DLA. Using TensorRT, you can offload supported layers onto DLA. We are adding support for more layers with efficiency. Please check [url]Developer Guide :: NVIDIA Deep Learning TensorRT Documentation for more details

3

Dear SivaRamaKrishna，

When I run the sample of “/usr/local/driveworks/bin/sample_drivenet”, for example
./sample_drivenet --precision=fp16
./sample_drivenet --dla=1 --dlaEngineNo=0 --precision=fp16
./sample_drivenet --dla=1 --dlaEngineNo=1 --precision=fp16
we can get a slower speed when we use DLA with FP16.
Is it normal or we need do some changes?

Thanks

4

So, in order to achieve better performance, I need to use GPU and NPU in parallel with driveworks2.0+cdua10.0+tensorrt5.0?

5

Dear denglong,
By default the application uses discrete GPU on DRIVE AGX Pegasus. Please set CUDA_VISIBLE_DEVICES=1 to select iGPU. DLA is another accelerator like GPU. Note that iGPU has more TOPS than DLA.