NUMBA, CUDA and QUTIP

Yash Tiwari

unread,

Oct 23, 2022, 12:13:14 AM10/23/22

to QuTiP: Quantum Toolbox in Python

Hello all,

I am trying to use GPU parallelization using different CUDA cores. Based on a quick read i found that NUMBA library is used for GPU parallelization .

https://numba.pydata.org/numba-doc/latest/cuda/overview.html

However I also came across a article which says that NUMBA works with only restricted set of packages like Numpy.

I want to ask does Qutip work with NUMBA ? Can I make use of NUMBA with Qutip to support GPU parallelization based on CUDA cores( on a single graphics card)

If not, which packages can be used with Qutip to support GPU parallelization based on CUDA cores( on a single graphics card).

Thanks.

Yash Tiwari

unread,

Oct 23, 2022, 12:14:18 AM10/23/22

to QuTiP: Quantum Toolbox in Python

+ However I also came across a article which says that NUMBA works with only restricted set of packages like Numpy.

https://stackoverflow.com/questions/50744686/numba-typingerror-cannot-determine-numba-type-of-class-builtin-function-or

Boxi Li

unread,

Oct 30, 2022, 5:35:18 AM10/30/22

to QuTiP: Quantum Toolbox in Python

Hello,

Unfortunately, it is not possible to use Numba directly with QuTiP. Numba supports many NumPy functions but qutip.Qobj it much more complicated than just a NumPy array.

For GPU support you can try to install manually:

- qutip-tensorflow https://github.com/qutip/qutip-tensorflow

- qutip-cupy https://github.com/qutip/qutip-cupy
- qutip-jax https://github.com/qutip/qutip-jax

They are all still under development, especially the JAX version. You can use them to perform common operations on Qobj, but none of them can be used directly with `mesolve` at this moment. You will have to find out an ODE solver yourself for the corresponding backend.