Some preliminary results suggest that 1D array operations are faster on CPU than on GPU if the number of elements is less than around 1e6.
Maybe GPUs will be more useful when we have higher modes if we can do some higher mode computation in parallel with matrix multiplications?
Some preliminary results suggest that 1D array operations are faster on CPU than on GPU if the number of elements is less than around 1e6.
Maybe GPUs will be more useful when we have higher modes if we can do some higher mode computation in parallel with matrix multiplications?