Insights on the SmaCS on-board AI-processor
August 6, 2021
The on-board AI-processor is intended to process very large data flows provided from the set of sensors

The SmaCS on-board AI-processor embeds an FPGA which performs signal pre-processing and up to three Jetson AGX Xavier SoCs from NVIDIA. These settings enable the system to handle up to 100 Gbit/s of video stream for a computing power of up to at 100 Tops and a consumption that ranges between 35 W and 120 W depending on the configurations.

The on-board FPGA displays 650K logical elements combined with 22MB of SRAM memory, up to three DDR4 memory blocks of 1GB each (at 2,666Mbps) and 512MB of NOR flash memory.

The Jetson AGX Xavier module with dimensions of 105 x 105 mm for a thermal envelope of less than 10 W, is deployed around a Xavier system chip. This SoC embeds a custom block of eight 64-bit general-purpose processor cores with ARM v8.2 architecture, a graphics processing unit (GPU) with 512 Volta cores and 64 Tensor cores, one or two deep learning algorithm accelerators (NVDLA), a VLIW machine vision accelerator, and one or two 4K / 60p / HEVC video encoders / decoders each with a computing capacity of 32 Tops.

more news