Maximum Qubit Support on GPU machine with different backend simulators


I wanted to know how many qubits at max can pennylane’s lightning qubit simulator support on GPU device?

Additionally, is there any detailed documentation on simulator backends and information on how many qubits each of it supports at maximum capacity?

Hello @kamzam, This post has been helpful: Pennylane lightning.gpu, 28+ qubits quantum circuit

1 Like

Hi @kamzam and @kevinkawchak ,

The number of qubits you can run depends on many factors. Here are some important factors:

  1. The compute resources you have. Running circuits on a laptop is more limiting than running them in a server or a supercomputer.
  2. The depth of your circuit. If you have a very shallow circuit you may be able to run larger circuits in some cases.
  3. The tools you use. If you’re using circuit optimization/compilation techniques like circuit cutting you might get to run circuits with more qubits.

If you’re running a circuit of moderate depth on a laptop without using circuit cutting or similar techniques, here’s a quick reference for how many qubits you could run using different simulators:

  • default.qubit and lightning.qubit: up to 26 qubits
  • lightning.gpu and lightning.kokkos (using GPUs): up to 29 qubits.
    If you optimize using MPI you can get maybe 3 extra qubits.
  • default.mixed: up to 16 qubits

Remember that aside from the device you will need to choose an interface, which is autograd/numpy for default but you can also choose Torch or Tensorflow.

You can dig deeper into choosing a device in this blog post.

I hope this helps you! Let us know how many qubits you get to run with each device.

Thank you @CatalinaAlbornoz, this was very helpful.

Thank you @kevinkawchak, I’ll go though this as well. :slight_smile: