Patened FFP8 for World-Beating Accuracy.
Hardware and software architecture deliver best-of-breed scalability.
Architecture-ensured low power and high performance accelerator for recommendation inferences.
NEUCHIPS RecAccel™ N3000 includes sophisticated hardwired accelerators, patented query scheduling, and a comprehensive software stack – all optimized for high accuracy, hardware utilization, and energy efficiency. Proprietary 8-bit coefficient quantization, calibration and hardware support deliver >99.95% of FP32 accuracy.
RecAccel™ N3000 employs the TSMC 7nm process, specifically designed for accelerating deep learning models. It exceeds 20M inferences per second at 20 Watts for leading recommender AI models.
The RecAccel™ SDK provides an automated path to run networks at high performance and accuracy with a streaming engine, configurable resource allocator, proprietary recommender operators, and calibrator.
The SDK toolkit also includes a Neuchips patented FFP Quantizer, bit-true simulator for offline simulation, a performance estimator, and a debugger for easy integration.