Jetson TX2 - Getting Started
Getting Started with Jetson TX2
NVIDIA Tegra SoCs contain a set of features that make them suitable for applications like inference on the edge for deep learning and artificial intelligence applications where the analysis and decision making needs to be done on-site as well as multimedia applications based in frameworks like GStreamer where capturing and encoding several camera streams is a must. Amongst the things that you need to know about Jetson TX2 are:
1) The Jetson TX2 Evaluation board is called the Jetson module as its predecessor TX1 and it can be bought directly from NVIDIA or Amazon.
2) TX2 SOMs can be bought individually for something around $300.
3) TX1 and TX2 are pin to pin compatible so all the TX1 carrier boards can be used with TX2 as well, however, TX2 has more reserved pins in use now.
4) It contains 6 processor cores organized into two clusters: ARM® Cortex® -A57 MPCore (Quad-Core) and NVIDIA Denver 2 (Dual-Core) Processor - 2GHz.
5) Questions can be posted on the devtalk forum where RidgeRun provides support and guidance
https://devtalk.nvidia.com/ - DavidSoto is my user.
6) Documents with very detailed information can be found in the Jetson download center
7) Jetson TX2 can encode 4K60fps H265 while Jetson TX1 did only 30fps. Now the encoder is fully accelerated by hardware, on TX1 the encoding was hybrid.
8) The codecs VP8, H264, H265, VP9 shown in figure 2 are accelerated by hardware and are available through GStreamer 1.0. Several instances can run at the same time. From NVIDIA datasheet:
9) Jetpack is the NVIDIA SDK and it basically flashes prebuild binaries into TX2, it also contains a scrip to download the code and build it manually which is a slow process. RidgeRun has an SDK that automates the process.
10) Jetpack 3.0 supports Jetson TX2 and one of the main differences is that it uses a much newer kernel 4.4 and installs ubuntu 16 on TX2. Also, it comes with GStreamer 1.8 and Linux 4 Tegra (L4T) 27.1
11) TX2 includes a GPU Pascal architecture instead of the NVIDIA Maxwell™ architecture. Both having 256 CUDA cores.
12) It contains 8GB of RAM
13) It supports up to 6 MIPI CSI cameras and you can capture them using GStreamer or libargus. RidgeRun creates drivers that are integrated with nvcamerasrc and therefore the ISP can be used to Debayer. TX2 supports 4 MCLK for cameras while TX1 only 2.
14) TX2 now adds two I2C ports more than TX1.
15) The display in TX2 supports GYNSC_HSYNC/VSYNC this is not available in TX1.
16) TX2 has two CAN controllers while TX1 didn't have any.
17) The Jetson EVM comes with the same Sony camera as TX1.
18) Jetpack 3.0 supports
Kernel version 4.4 Multimedia API 27.1 OpenGL 4.5 OpenGL ES 3.2 EGL 1.4 CUDA 8.0.64 libargus 0.95 X11 Support U-Boot
19) The table shows the main differences can be found in Linux Jetson wiki page
I2C
TX2 now adds two i2c ports more than TX1. So in total, it has 9 I2C buses.
ls /dev/i2c-* /dev/i2c-0 /dev/i2c-2 /dev/i2c-4 /dev/i2c-6 /dev/i2c-8 /dev/i2c-1 /dev/i2c-3 /dev/i2c-5 /dev/i2c-7
i2c interfaces are defined in: hardware/nvidia/soc/t18x/kernel-dts/tegra186-soc/tegra186-soc-i2c.dtsi
Below is a table that summarizes the I2C Bus mapping on TX2:
| I2C bus number | I2C bus address |
|---|---|
| i2c-0 | i2c@3160000 |
| i2c-1 | i2c@c240000 |
| i2c-2 | i2c@3180000 |
| i2c-3 | i2c@3190000 |
| i2c-4 | i2c@31a0000 |
| i2c-5 | i2c@31b0000 |
| i2c-6 | i2c@31c0000 |
| i2c-7 | i2c@c250000 |
| i2c-8 | i2c@31e0000 |
