> exact reproducibility between simulations, even within the same >> environment, is not guaranteed for MCX, and accumulation operations on the >> GPU in general

that exact reproducibility of any output > quantity that are accumulated over multiple threads, such as > fluence, is not guaranteed in mcx. Even using the same seed, > running

"exact reproducibility is not guaranteed" is for >>> the output fluence - because accumulation of floating point numbers from >>> multi-threads

"exact reproducibility is not guaranteed" is for the output fluence - because accumulation of floating point numbers from multi-threads may produce different outputs

Kris exact reproducibility has not been a guarantee on the GPU, it just happened that NVIDIA GPUs gives "nearly" repeatable results (likely due to a stable thread scheduling

excellent > reproducibility due to some fixed thread scheduling method, so mcx has been > able to generate reproducible results. In comparison, MMC is not able to > produce

concerning the reproducibility of simulations. I observed that running the same simulation (with identical parameters) leads to different results (measured as the sum of all output