Nvidia Reveals World’s Smallest AI ‘Supercomputer’

Article By : Sally Ward-Foxton

The board delivers 1.4 TOPS/W for drones and small robots.

Nvidia has launched the Jetson Xavier NX, a new board which features the computing power of the AGX Xavier in the form factor of the Jetson Nano (70 x 45mm). The Xavier NX board can achieve 14 to 21 TOPS INT8 performance at power budgets between 10W and 15W, respectively. This amount of processing power would enable running multiple neural networks in parallel, and processing data from multiple high-resolution image sensors simultaneously, according to the company.

The Jetson Xavier NX board features an Nvidia Volta GPU AI accelerator with 384 CUDA cores and 48 tensor cores, with two Nvidia deep learning accelerators. Its CPU is a 6-core Nvidia Carmel Arm 64-bit CPU.

The Jetson NX joins the Jetson AGX Xavier (up to 32 TOPS) for embedded applications including robotics, and the Jetson Nano, a small form factor embedded board that offers 0.5 TFLOPS of FP16 performance. There is also the Jetson TX2 – the NX is smaller and delivers significantly higher (up to 15x) performance in the same power bracket.

Nvidia Jetson Xavier NX

Nvidia’s Jetson Xavier NX board is smaller than a credit card and provides 1.4 TOPS/W (Image: Nvidia)

“Jetson Xavier NX fits a needed market space between TX2 and AGX Xavier,” said Patrick Moorhead, president and principal analyst, Moor Insights and Strategy. “It’s ideal for devices with the need for many sensors, even a battery-powered, highest-performance drone. The biggest benefit is a massive amount of ML inference performance with a mature toolchain… There’s nothing out there like it with that much ML performance at that power envelope.”

“This is the world’s smallest AI supercomputer,” said Rob Csongor, VP of Autonomous Machines at Nvidia. “It’s Xavier-level performance in a Nano size.”

The applications Nvidia is targeting include embedded and robotics applications with 10W to 15W of power available. Applications include small commercial delivery robots and drones which need to process multiple camera streams, robot arms which include high resolution sensors in the logistics and production markets, video analytics systems where space is constrained, and handheld imaging devices such as medical devices.

“There are very important specific segments of the market that are defined by a need for increasing performance, in a box that does not increase in size,” Csongor said. “The power budget also does not increase. And as a result, if you are not able to deliver a lot or performance within those constraints, then the applications are constrained. Xavier NX is perfect for these segments.”

The Jetson Xavier NX, which is pin-compatible with Jetson Nano, will be available in March 2020.

Leave a comment