YPOLOGIST
Green Novel Architecture
A p-Cell Programmable, General-Purpose, Parallel Accelerator for Heterogeneous Systems with applications in LLM and video pattern recognition (YOLO)
Main functions accelerated: matrix multiplication (GeMM) and convolution.
Achieves 2-10x energy savings in linear algebra applications compared to off-the-shelf solutions
Delivers 337% better performance in MH/sec/Watt for blockchain applications compared to NVIDIA GPUs
Provides up to 4px speed improvement in ASIC implementations compared to mono-core solutions
11x
Energy Efficiency in ML
3x
Less Area
6px
Acceleration
>3x
Less Cost
*The comparisons are made with the Nvidia chip implemented on the same technology, operating at the same frequency.