Nvidia equips Jetson Xavier NX module with Volta GPU with 384 cuda cores

Nvidia presents the Jetson Xavier NX, a module for use in autonomous robots or other systems. The new module will have a Volta GPU with 384 cuda cores and 48 Tensor cores. The Xavier NX is pin compatible with the previously released Jetson Nano.

The new Jetson module has a Xavier soc with six Nvidia Carmel cores. These are 64bit cores based on the ARMv8.2 architecture with a clock speed of 1400MHz. When using a maximum of two cores, that speed can be increased to 1900MHz. The soc has 6MB L2 cache and 4MB L3 cache.

The soc has a Volta GPU with 384 cuda cores at 1100MHz and 48 Tensor cores for machine learning. Furthermore, there is 8GB lpddr4x memory on the module, with a memory bus of 128bit and a speed of 1600MHz, which provides a bandwidth of 51.2GB/s. There is also 16GB of emmc flash storage. Twelve MIPI CS-I2 lanes are available for controlling cameras.

Manufacturers can set the module to a 10W or 15W mode. According to Nvidia, that yields 14tops and 21tops respectively. In 10W mode, it uses up to four CPU cores and clocks the GPU down to 800MHz.

The Jetson Xavier NX module measures 45×69.6mm and features the same 260pin so-dimm connector as the less powerful Jetson Nano that Nvidia announced early this year. That version has a Maxwell GPU with 128 cuda cores. The new model has much more computing power, but is also considerably more expensive. Nvidia will release the new model in March next year for $399. The Jetson Nano retails for $129.

Nvidia is targeting manufacturers of autonomous machines with the Jetson Xavier NX module. According to the manufacturer, the module is suitable for simultaneously running different neural networks and processing images from cameras. The new Xavier NX is a smaller variant of the Jetson AGX Xavier with 512 cuda cores, which was released earlier this year.