Nvidia presents Jetson AGX Orin module for robots and AI applications

Nvidia introduces a new version of its compact Jetson module for AI applications. The Jetson AGX Orin has an Ampere GPU with 2048 CUDA cores and an Arm CPU with 12 cores. The system can perform up to 200 trillion calculations per second.

With a peak performance of 200Tops in int8 calculations, the Jetson AGX Origin is up to six times faster than its predecessor, the Jetson AGX Xavier, Nvidia says. Incidentally, this only applies to these specific calculations with a low precision, which are used in neural networks.

Predecessor Jetson AGX Xavier had a Volta GPU with 512 cores, the new model has an Ampere GPU with not only 2048 CUDA cores, but also 64 Tensor cores. The system is equipped with 32GB of LPDDR5 memory with a throughput of 204.8GB/s.

Nvidia ships the Jetson AGX Orin as a module and a developer kit. The separate board has dimensions of 100x87mm and is intended for integration into robots or other equipment that must be able to function autonomously. The manufacturer cites examples such as systems in factories, robots for logistics and large industrial drones. The module consumes 15W to 50W, depending on the setting. Nvidia has published more technical details in a PDF.

Nvidia will release the Jetson AGX Orin in the first quarter of 2022. The manufacturer does not mention a price.

Medical applications

Nvidia also introduces Clara Holoscan, a computing platform for the medical industry that works with the Jetson AGX Orin module. According to Nvidia, that platform is suitable for real-time processing of information from medical devices.