NVIDIA Jetson Embedded Computers

Originally named the Tegra X2, the Jetson TX2 is the oldest of the Jetson Range. It features NVIDIAs own microarchitecture (code-named Denver2) — a 64/32bit multi-core ARM v8 processor and a 256-core Nvidia Pascal GPU, up to 8GB 128-bit DDR4 onboard memory and up to 32GB onboard eMMC storage. The Jetson TX2's 15W module has been used in projects far and wide, from military drones processing high-resolution image feeds in real-time with object classification, to monitoring the health of beehives by simultaneously analysing audio, video and sensor data streams. 

The Jetson TX2 currently comes in 3 main models: a 4GB or 8GB TX2, and the more recent Jetson TX2 NX which has the same specifications as the TX2 4GB but uses the 260-pin SODIMM form factor of the Jetson Nano. The Jetson TX2 NX can also be swapped into the carrier board of the Nano to scale-up performance.

The Jetson Xavier NX is the same form factor as the Jetson nano and TX2 NX, but boasts a huge bump in performance without increasing the power consumed. The Xavier NX SoM leverages NVIDIA's latest ArmV8.2 64-Bit microarchitecture (code-named Carmel), a Hexa-core processor and a 384-core Nvidia Volta GPU. Unlike the Nano and the TX2, there is some additional AI focused processing power, through 48 Tensor cores and dual NVDLA Engines. As with the other Jetson SoMs, there is 8GB LPDDR4X memory and 16GB eMMC storage onboard.

From mobile robots and drones to portable medical equipment and embedded IoT systems, the Jetson Xavier NX is suitable for autonomous systems and intelligent edge devices that require high-performance AI and complex Deep Neural Networks in a compact, low-power form factor.

The Jetson AGX Xavier is the largest of the Jetson modules, offering up 32 TOPS of INT8 performance. Like the Xavier NX it is built around the ARM V8.2 64-bit Carmel microarchitecture, but this time with an Octa-core processor, a 512-Core Volta GPU, 64 Tensor cores and dual NVDLA engines. There is also 32GB of 256-bit LPDDR4x onboard memory and 32GB of onboard eMMC storage included. All this extra processing power targets AGX Xavier a bit differently — providing this exceptional GPU performance in a 30W package means the AGX Xavier is capable of running autonomous mobile robots (AMRs) or being used for inference for large scale machine vision at the edge.

All of the Jetson boards have a small amount of PCIe support for the addition of a capture card or extra I/O. The AGX Xavier can support multiple expansion devices through its PCIe Gen4.0 controller, including simultaneous support for multiple PCIe devices such as an additional GPU through PCIe x8.