Today, NVIDIA is expanding on the last segment of its business by teasing Xavier. Xavier, which was announced at the European edition of the GPU Technology Conference, is an artificial intelligence (AI) supercomputer that was designed from the ground-up to be used for self-driving vehicles.
Xavier is an SoC built around a custom 8-core processor architecture, and is backed by an all-new Volta GPU design. Xavier is being billed as a replacement for the recently announced Parker-based Drive PX 2 platform (which is powered by two NVIDIA Denver cores and 4 ARM Cortex-A57 cores) that is more power efficient and provides much of its computational power in a single SoC.
Whereas Parker features Pascal GPU architecture with 256 CUDA cores, Xavier’s Volta raises the stakes with 512 CUDA cores. Xavier’s video processing engine is also capable of 8K encode/decode (7680x4320) compared to 4K (3840x2160) for Drive PX 2.
NVIDIA says that Drive PX 2 offers 24 DL TOPS (Deep Learning Tera-Ops) of single-precision compute performance with a TDP of 80 watts. Xavier counters with 20 DL TOPS while operating within a TDP of 20 watts. That is a significant power savings, especially considering that Drive PX 2 and Xavier are both built using a 16nm FinFET process.
“This is the greatest SoC endeavor I have ever known, and we have been building chips for a very long time,” said NVIDIA CEO Jen-Hsun Huang when announcing Xavier.
If you’re wondering why NVIDIA would announce Xavier now so soon after Drive PX 2, it’s because it won’t even begin sampling until Q4 2017. Volume production isn’t expected until 2018.