A Hardware Efficient Random Number Generator for Nonuniform Distributions with Arbitrary Precision

Table 1

Normal random number generator architecture comparison.

Implementation	Architecture	Power consumption	Throughput [M samples/𝑠]	Energy per sample
Fast Mersenne Twister,; optimized for SIMD	Intel Core 2 Duo PC 2.0 GHz,; 3 GB RAM, one core only	~100 W	600	166.67 pJ
Nvidia Mersenne Twister + Box-Muller CUDA	Nvidia GeForce 9800 GT	~105 W	1510	69.54 pJ
Nvidia Mersenne Twister + Box-Muller OpenCL			1463	71.77 pJ
Proposed architecture, only one instance [1]	Xilinx FPGA Virtex-5FX70T-3 380 MHz	~1.3 W	397	3.43 pJ
Proposed architecture, 100 instances		~1.9 W	39700	0.05 pJ