Can we have DeepStream SDK on Jetson TX-1?

[burakmandira]

Hi,

I have two questions:

1. From the response of my application to DeepStream SDK (1.0) on Jetson, I know that SDK 1.0 is under progress till the next version is developed. Also, in https://devtalk.nvidia.com/default/topic/1025903/announcements/nvidia-deepstream-sdk-on-jetson-early-access-program/#reply this page, it’s said that:

@dusty_nv

Hi lakshmih85, this SDK is still under development and general availability is sometime later this year. We appreciate your interest and will pass this to the team. In general, early access programs have limited slots and the team is trying to expand the availability as fast as possible.

I have downloaded the latest version SDK 1.5 and in its manual pdf, it’s stated that NVIDIA recommends that DeepStream be run on a hardware platform with an NVIDIA Tesla® P4 or P40 graphics card. Is this just a recommendation or a prerequisite system requirement? i.e. can we deploy SDK 1.5 on Jetson TX1?

2. In GitHub page of Jetson Inference ( https://github.com/dusty-nv/jetson-inference ), there exist source code where compilation from it explained in detail. Can we use that source code instead of downloading DeepStream SDK 1.0 (which is currently unavailable to download) from Nvidia’s website?

Waiting for your response.
Thanks.

[dusty_nv]

I have downloaded the latest version SDK 1.5 and in its manual pdf, it’s stated that NVIDIA recommends that DeepStream be run on a hardware platform with an NVIDIA Tesla® P4 or P40 graphics card. Is this just a recommendation or a prerequisite system requirement? i.e. can we deploy SDK 1.5 on Jetson TX1?

Hi burakmandira, this is referring to “DeepStream for Tesla”, there is also a version in development “DeepStream for Jetson”. Both have the same shared API for portability across the platforms, however the binaries for the Tesla version won’t run on Jetson.

2. In GitHub page of Jetson Inference ( https://github.com/dusty-nv/jetson-inference ), there exist source code where compilation from it explained in detail. Can we use that source code instead of downloading DeepStream SDK 1.0 (which is currently unavailable to download) from Nvidia’s website?

You can certainly follow the jetson-inference repo to get started training & deploying your own deep learning networks, the tutorial highlights the DIGITS->TensorRT workflow. jetson-inference is meant mostly for deployment on Jetson, although some folks on GitHub have it building and running on x86 with discrete GPU (unofficially).