Fig. 8From: Heterogeneous computing for epidemiological model fitting and simulationGiven that batches of particles can be evaluated asynchronously, one can ship off multiple batches of particles to the GPU in parallel. Once a CPU thread ships off a batch of particles to the GPU, it needs to wait for it to be finished before continuing to work. Using fibers, a CPU thread can process multiple batches at the same time and therefore increases performance. The x-axis depicts the amount of fibers per CPU thread and the y-axis depicts the speedup of the GPU compared to the best CPU timing. A total of 2048 particles were used with 64 particles per batchBack to article page