Open hardware specialist BeagleBoard.org has unveiled a new entry in the BeagleBone family of single-board computers: the BeagleBone AI-64, its first 64-bit design tailored for embedded artificial intelligence (AI) and machine learning (ML) workloads.
"BeagleBone AI-64 represents a major milestone for BeagleBoard.org, satisfying some of the most requested features from our developer community," says Jason Kridner, BeagleBoard.org Foundation board president of the latest design, "including 64-bit support and inclusion of PCIe [PCI Express] on an expansion header. With multiple SuperSpeed USB ports, familiar BeagleBone cape expansion headers, and desktop-capable performance, the general-purpose embedded applications for this board are endless, with 8 TOPS neural-network performance accessible through familiar Python libraries to boot."
The board, which is designed for compatibility with both existing BeagleBone Cape add-ons and mikroBUS Click boards, is driven by a Texas Instruments TDA4VM system-on-chip with two 64-bit Arm Cortex-A72 cores running at up to 2GHz, a C7x floating-point vector-accelerating digital signal processor (DSP) running at up to 1GHz with a claimed 80 GFLOPS/256 GOPS of compute, two C66 floating-point DSPs with an additional 40 GFLOPS/160 GOPS at up to 1.35GHz, a deep-learning matrix multiply accelerator (MMA) coprocessor offering a claimed 8 TOPS of INT-8 precision compute at 1GHz, vision processing accelerators (VPACs), depth and motion processing accelerators (DMPACs), six Arm Cortex-R5F microcontroller cores running at up to 1GHz, a PowerVR Rogue 8XE GE8430 3D-capable graphics processor running at up to 750MHz with 96 GFLOPS of compute and up to six gigapixels per second throughput, and twelve multi-channel audio serial port (MCASP) modules.
In other words, the BeagleBone AI-64 is a bit of a beast. To this, the designers have added 4GB of LPDDR4 memory, 16GB of eMMC flash with a microSD slot for expansion, a single USB 3.0 Type-C port and two USB 3.0 Type-A ports all offering USB SuperSpeed support for 5Gb/s throughput, a gigabit Ethernet port, a mini-DisplayPort connector, two four-lane CSI camera connectors, a single four-lane DSI display connector, and an M.2 PCI Express E-key connector designed for optional Wi-Fi, Bluetooth, or cellular communications boards.
Naturally, the board's design is open hardware — and its toolchain is also open source. Its designers say that the BeagleBoard AI-64 is ideally suited for everyone from hobbyists looking for a ready-to-roll platform for experimentation with computer vision, machine learning, robotics, and more to engineers seeking to take the central design and customize it for their own needs.
"We believe this board will capture the imaginations of designers and empower them to build complete and powerful AI systems," says Christine Long, BeagleBoard.org Foundation chief executive. "At [its] extremely competitive price point, we are excited about the new applications that BeagleBone AI-64 will enable for new and experienced users."