Talos is a custom FPGA-based hardware accelerator built from the ground up to execute Convolutional Neural Networks with extreme efficiency. It isn't just a reimplementation of existing software logic in hardware; it is a rethinking of how deep learning inference should work at the circuit level.
Most deep learning frameworks are built for flexibility. They handle dynamic graphs, varying batch sizes, and a multitude of layer types. Talos takes the opposite approach. It strips away the runtime, the scheduler, and the operating system overhead to expose the raw compute capability of the FPGA. By implementing the entire inference pipeline in SystemVerilog, we achieve deterministic, cycle-accurate control over every calculation.
Most deep learning frameworks are built for flexibility. They handle dynamic graphs, varying batch sizes, and a multitude of layer types. Talos takes the opposite approach. It strips away the runtime, the scheduler, and the operating system overhead to expose the raw compute capability of the FPGA. By implementing the entire inference pipeline in SystemVerilog, we achieve deterministic, cycle-accurate control over every calculation.