Radical Solution for AI Acceleration; No Model Performance Sacrifices

Today's general-computing systems, designed to run multiple programs simultaneously, create bottlenecks for AI models. Resource contention cause exaggerated latency, dropped frames, and extreme power consumption.

Models powered by GigaMACS™ do not replicate the bottlenecks. Frame grabbers, which add to latency, are eliminated since the video stream is accepted directly from the camera. Eliminating RAM and using dedicated resources reduces latency to near-zero producing outputs in real-time. With the computational effort reduced to an absolute minimum, 100% of the frames are processed without any lost data. GigaMACS™ on FPGAs and ASICs have minimal resource overhead and use 90% less power than GPUs and TPUs.

GigaMACS™ accelerates artificial intelligence to accept and process live video streams with high-resolution frames faster than any GPU or TPU available today and uses only a tenth of the power while constantly maintaining near-zero latency.

The pipelined circuitry is a radical design that completely replaces the von Neumann architecture.