Frequently Asked Questions

I do not have an NVIDIA GPU on my system. Which version should I install ?

You should use the CPU version. Install (CPU-only)

I have an entry-level GPU, but my circuits are usually small with fewer qubits and gate operations. Which version should I install ?

You should use the CPU version. Install (CPU-only)

I have an entry-level GPU with 4GB of memory. My circuits are complex. Which engine should I use?

You should install the GPU version. You may benefit from selecting the hybrid engine “serin_quantum_engine” when you acquire the backend.

Why is the GPU version slower than the CPU version on my system?

It takes a while to set up the GPU kernels. If you are using a quantum circuit with fewer gates or implementing simple circuits like 2-local circuits, or GHZ states, you can use the CPU version.

The GPU-enabled SDK throws errors when imported in Python. What is the solution?

Please see Install (GPU-enabled). You have to set the path to the CUDA Toolkit. If you are using Windows, you have to set the path to the CUDA Toolkit in Python.

Which is the preferred way of storing the account name and account key?

You can provide the account name and the access token explicitly as arguments to the QuantumRingsProvider function, store them as environment variables and fetch them from the environment at runtime, or save them in the local configuration file using the save_accounts API.

How do I change the precision?

You can select the precision – single or double – when acquiring the backend or while issuing the run command. The default is “single” precision unless it is changed.

Why does get_counts() return {} (an empty dictionary)?

Execution may be asynchronous: backend.run(...) can return before sampling is finished. If you call job.result() too soon, get_counts() may not have data yet.

For minimal examples, pass mode="sync" to backend.run(...) so the call blocks until the job completes. For asynchronous execution, wait for completion using job_monitor or wait_for_final_state before calling job.result(). See Run Settings.