NVIDIA Launches the IGX Orin Developer Kit, Aiming for Industrial and Medical Edge AI Workloads
Featuring a dedicated safety processor and 248 TOPS of compute, this beastly box could be driving future robo-surgeons.
NVIDIA has announced the launch of a non-Jetson edge AI development kit based on a IGX Orin system-on-module — and aimed very firmly at industrial and medical environments.
"Imagine you are about to embark on the mountain biking adventure of a lifetime. You have done all of the planning and training. Now, all you need is the perfect bike. You need the best shocks, tires, brakes, frame, handlebars, and seat. Imagine that all of these parts would come together in one package, pre-assembled, saving you time and money," NVIDIA's Grayson Davis says of the company's latest launch, which is not a bike. "NVIDIA IGX Orin offers a similar package to edge AI application developers: an optimized, all-in-one platform. NVIDIA IGX Orin is the first platform to combine industrial-grade hardware with enterprise-level software and support for edge AI management."
The kit is built, as the name implies, around the NVIDIA IGX Orin system-on-module (SOM), which sits outside the Jetson ecosystem. The SOM includes a 12-core Arm Cortex-A78AE v8.2 processor running at up to 1.996GHz, an Ampere-architecture graphics processor with 2,048 CUDA cores and 64 Tensor cores running at up to 1.185GHz, two NVIDIA Deep Learning Accelerator 2.0 (NVDLA 2.0) coprocessors running at up to 1.4GHz, and one Programmable Vision Accelerator v2 (PVA v2), plus 64GB of LPDDR5 memory with 204.8GB/s bandwidth and 64GB eMMC storage.
Those specifications put the SOM on a level playing field with the Jetson AGX Orin 64GB SOM, though NVIDIA is oddly claiming just 248 tera-operation per second (TOPS) for the IGX compared with 275 TOPS for the AGX. It's the kit's other capabilities which make it stand out, including an optional NVIDIA A6000 GPU add-in board for additional compute performance and a ConnectX-7 smart network interface card (SmartNIC) which can push an impressive 200Gb/s throughput.
Elsewhere on the board are hardware safety extensions, a functional safety island (FSI), and a dedicated Infineon Aurix TC397 safety microcontroller unit — all of which peg the kit as being targeted firmly at industrial users. Indeed, NVIDIA's suggested use-cases include edge AI projects from proactive safety measures like camera-based barrier breach detection and collision detection to predictive maintenance, field inspections, and medical tasks including AI-powered endoscopy, ultrasound, and even robotic surgery.
One thing NVIDIA isn't sharing publicly, however, is the price — but interested parties can apply for pricing information through the official product page.