Simulation precision
Qibo allows the user to easily switch between single (complex64)
and double (complex128) precision as described in
How to modify the simulation precision?.
In this section we compare simulation performance of both precisions.
We find that as the number of qubits grows using single precision is ~2
times faster on GPU and ~1.5 faster on CPU.
nqubits |
CPU c64 | CPU c128 | GPU c64 | GPU c128 |
|---|---|---|---|---|
| 23 | 0.41793 | 0.57448 | 0.07276 | 0.09249 |
| 24 | 0.83167 | 1.28207 | 0.09982 | 0.14664 |
| 25 | 1.75593 | 2.68645 | 0.15088 | 0.25463 |
| 26 | 3.68007 | 5.60514 | 0.26269 | 0.47597 |
| 27 | 7.71366 | 11.78381 | 0.48103 | 0.93437 |
| 28 | 16.36245 | 24.84334 | 0.93884 | 1.87903 |
| 29 | 34.66599 | 52.47754 | 1.88652 | 3.8849 |
| 30 | 73.43444 | 110.6095 | 3.90816 | 8.10977 |
| 31 | 155.01974 | 232.35723 | 8.08614 | |
| 32 | 329.56916 | 488.00321 | ||
| 33 | 694.67844 | 1021.06619 | ||
| 34 | 1465.81645 |
