Skip to main content
. 2016 Jul 21;10:333. doi: 10.3389/fnins.2016.00333

Table 1.

Summary of comparison of various RPU system designs and state-of-the-art CPU and GPU.

System Throughput (TeraOps/s) Power (W) Power efficiency (GigaOps/s/W) Network size (number of weights) Acceleration vs. CPU
CPU Power8 12 Cores 0.676 250 2.7 1
GPU NVidia Tesla K40 4.3 242 17.8 6.4
Design 1 5000 250 20,100 200 M 7400
Design 2 21,000 250 83,800 840 M 31,000
Design 3 420 22 19,000 1680 M 620