CUDA-Q Backend Configuration

The CudaQBackend is highly optimized to run directly on the NVIDIA CUDA-Q platform, supporting multi-node, multi-GPU frameworks efficiently.

Installation

Ensure you have the required extra installed:

pip install hlquantum[cudaq]

This installs the cuda-quantum Python Wheel locally.

Target Selection

By default, NVIDIA CUDA-Q will use the fast qpp CPU simulator. When you provide an NVIDIA GPU, you can leverage native CUDA targets capable of distributing quantum state arrays across distributed memory architectures.

You configure your CUDA-Q setup by passing explicit "nvidia", "nvidia-fp64", or "nvidia-mgpu" text markers to the CUDA-Q target backend wrapper.

from hlquantum.backends import CudaQBackend
import hlquantum as hlq

# Instantiate an NVIDIA multi-GPU node backend
backend = CudaQBackend(target="nvidia-mgpu")
result = hlq.run(circuit, backend=backend, shots=1000)

No external API keys are required for NVIDIA simulators. When working with NVIDIA systems via cloud virtual environments or Docker nodes containing multiple devices, ensure variables like CUDA_VISIBLE_DEVICES remain exported natively into the terminal context as standard. Or, utilize HLQuantum GPUConfig(enabled=True) across your instance.