Nvidia announced an “Isaac” software developer platform for robots and other autonomous machines that runs on its Linux-friendly octa-core, ARM64 “Jetson Xavier” module with integrated high-end Volta GPU. A $1,300 dev kit is due in August.

In 2016, Nvidia unveiled its AI-focused Nvidia Xavier computer-on-module, a more powerful follow-on to its Tegra-based Jetson modules. Last October, the company announced a Drive PX Pegasus autonomous car computer board that incorporates up to 4x Linux-driven Xavier modules, each with high-end 512-core Nvidia Volta GPUs, plus two more discrete GPUs. Now, the company is deploying Xavier on a dev board that will power robots, drones, and other autonomous machines. The key ingredient is a new Nvidia Isaac software platform that includes an SDK, IMX algorithms, and a Sim virtual simulation environment.

Nvidia Xavier (left) and Isaac IMX data points for visual recognition
(click images to enlarge)

The upcoming, $1,299 Nvidia Jetson Xavier developer kit will target applications that require “maximum compute at the edge to run modern AI workloads and solve problems in manufacturing, logistics, retail, service, agriculture and more,” says Nvidia. The company did not provide details about the kit, which includes the Isaac software, and will also presumably include a development board with one or more Xavier modules.

Nvidia made no mention of OS support. As noted, however, the Xavier runs Linux on the Drive PX Pegasus, and in April the module received support in the Linux 4.17 kernel.

Nvidia Isaac architecture
(click image to enlarge)

The Isaac robotics software consists of:

• Isaac SDK — a collection of APIs and tools to develop robotics algorithm software and runtime framework with fully accelerated libraries
• Isaac IMX — Isaac Intelligent Machine Acceleration applications, a collection of NVIDIA-developed robotics algorithm software
• Isaac Sim — a highly realistic virtual simulation environment for developers to train autonomous machines and perform hardware-in-the-loop testing with Jetson Xavier

 
Nvidia Xavier

The large, 100 x 87mm Nvidia Xavier module, which is also known as the Tegra194, has 20 times the performance and 10 times the energy efficiency of the Jetson TX2 module, claims Nvidia. It comprises 9 billion transistors that help to deliver 30 TOPS (trillions of operations per second) of performance. The version included in the Pegasus, which was said to integrate 7 billion transistors, consumed only 30 Watts.

The Xavier runs on 8x ARMv8.2 CPU cores with 8MB L2 and 4MB L3 cache, along with a Volta GPU with 512 Tensor Cores. There are also a pair of NVDLA accelerator engines and a 7-way VLIW vision accelerator co-processor. The module can generate dual 4Kp60 video encoding with HEVC and dual 4Kp60 video decoding with 12-bit support.

Drive PX Pegasus

The module is further equipped with 16GB LPDDR4 RAM and 32GB eMMC 5.1. The 699-pin board-to-board connector can drive a carrier board with a GbE port, 5x 16GT/s Gen 4 PCIe controllers, and HDMI, DP, and eDP display interfaces.

The Xavier supplies 16 lanes of CSI-2 and 8x lanes of SLVS-EC for up to 16 simultaneous cameras. You also get support for 3x USB 3.1 and 4x USB 2.0 ports, as well as UFS, I2S, I2C, SPI, CAN, GPIO, UART, and SD interfaces.

 
Further information

The $1,299 Nvidia Jetson Xavier developer kit will be available to early access customers starting in August. More information may be found in Nvidia’s Isaac announcement and Isaac product page, as well as the Xavier product page.