Graphics big NVIDIA has not given up on the development of the central processor, but based on the ARM reduced instruction set, after the first two auto-driving platforms Drive PX/PX2, the Xavier Soc released this year at CES is the new answer for NVDIA. .
This platform is not only used for auto driving, but also for robots with continuous learning ability.
NVIDIA announced the main information of the Jetson Xavier development platform on the official Youtube channel on July 10, including parameters, appearance, interface, performance and more.
Jetson Xavier consists of two modules, the calculation block (three measurements 100x87x16mm) and the carrier board integrated with the Xavier chip. The carrier board has various I/O interfaces and pre-installed heat sinks.
The Xavier compute module and the carrier board are connected via a 699Pin mezzanine connector. Currently, the PCIe4.0 specification is supported. If you remember correctly, this is NVIDIA's first PCIe 4.0 product.
In terms of performance, Xavier SoC is based on TSMC's 12nm process, integrating 9 billion transistors, chip area of 350 square millimeters, CPU using NVIDIA self-developed 8-core ARM64 architecture (codenamed Carmel), GPU with 512 CUDA Volta, support FP32/FP16/INT8 , 20W power consumption single-precision floating-point performance 1.3TFLOPS, Tensor core performance 20TOPs, unlocked to 30W after 30W.
Jetson Xavier is scheduled to go on sale in August for $1,299.
