![]() ![]() From what I understand, Prime95 is currently trying to find 10+ million digit primes, and is actually reaching the limit on the size of the number it can calculate using 64 bit precision (its peak is around 23 million digits.)įrom what I understand, 8800’s only operate on 32 bit floats, meaning it could be ported, but the number of digits that could be calculated within that smaller precision would be lowered significantly. Prime95 uses a discrete weighted fourier transform, which is currently implemented using 64 bit floating point arithmetic. There may be other extended precision representations which are more efficient. Adding two pseudo-double precision numbers using the dsfun90 method takes 11 single precision additions, and multiplication takes somewhere around 25 operations since the G80 can do a single-precision mutliply-add in one instruction. Of course, it isn’t as fast as dedicated silicon would be. Download Prime95 - a handy tool for overclockers and system stability checkers, Prime95 has a feature called 'Torture Test' that allows maximum stress testing on the CPU and RAM. ![]() (I recently had to port the addition and multiplication subroutines from dsfun90 to CUDA so I could implement an exponential with 10^-12 precision using the 10^-7 precision floats.) The dsfun90 library is one example of how to do this. You can implement higher precision arithmetic with lower precision floats.
0 Comments
Leave a Reply. |