dantegd
commented
1 month ago Thanks for the issue @JiayuChen02, I just tried your code on a machine with a T4 and this is the results I got:
{'alpha': 0.001, 'copy_X': True, 'fit_intercept': True, 'max_iter': 1000, 'positive': False, 'precompute': False, 'random_state': None, 'selection': 'cyclic', 'tol': 0.001, 'warm_start': False}
scikit-learn training time: 7.7504 seconds
{'output_type': 'input', 'verbose': 4, 'handle': <pylibraft.common.handle.Handle object at 0x7fd8addf73f0>, 'fit_intercept': True, 'normalize': False, 'tol': 0.001, 'solver': 'cd', 'max_iter': 1000, 'alpha': 0.001, 'selection': 'cyclic'}
cuML GPU training time: 0.8425 secondsIn general, 300 rows is not a dataset where I would expect massive GPU speedups, but things seem to be working fine. What could be happening is you could be running maybe into some JIT compilation issues? What GPU are you using? It would be very helpful if you could run this script and put the output here: https://github.com/rapidsai/cuml/blob/branch-24.10/print_env.sh
JiayuChen02
I tried to compare the speed difference of the same model on CPU and GPU. Here is my code：
And the result is
Why?