Precision Of Parameters & Energy

Only float and double are supported. Default is float.

UFSACL::UltraFastSimulatedAnnealing<106*2, 100000,double> simHighPrecision(" energy += 1.0;");
UFSACL::UltraFastSimulatedAnnealing<106*2, 100000> sim(" energy += 1.0f;");

Desktop/gaming GPUs have very low 64bit fp performance. CPUs have generally 1/2 ratio of 64/32 bit performance ratio:

"Fitting 106 circles in a square with 10x width" sample with double-precision run:

OpenCL device info:
GeForce GT 1030 computed 6.103% of total work
gfx1036 computed 4.155% of total work
AMD Ryzen 9 7900 12-Core Processor computed 89.742% of total work

Two GPUs with 512 combined shaders have only 1/9 of performance of a CPU with 12 cores.