YPOLOGIST - Capabilities Statement


Ypologist's XPU is a massively parallel novel architecture targeting Deep Learning and Neural Network applications.
XPU's many-core FPGA and ASIC implementations will deliver excellent cost and power-efficient alternatives to accelerators such as Nvidia's GPCPU or Intel's MIC.

Performance highlights

The main application domains, for which libraries will be developed, are linear algebra, cryptocurrencies, automotive and bio-informatics. The following are the performance characteristics for each of these applications:

  • Linear Algebra 2 to 3X improvement in energy savings and 3X increase the ratio performance/peak performance compared with off-the-shelf solutions. In most Big-Data applications power consumption per task is 20%.
  • Cryptocurrency applications request high MH/sec/Watt (MH stands for mega hashes). Measured performance 337% against NVIDIA GPU.
  • In automotive stereo vision, where tight real-time performance and temperature requirements are mandatory, XPU's 1024-cell system consumes less than 6W at 100 C.
  • Bio-Informatics applications are accelerated 6X when implemented in a 28nm FPGA when compared with a 22nm 4-core Intel CPU with SSE. Simulations predict a 30X improvement. The energy efficiency is improved 20X for the XPU FPGA version and 300X for the XPU ASIC version.