Q-GPU: Quantum Circuit Simulation Using GPUs
A University of Pittsburgh researcher has designed a novel approach (Q-GPU) to optimize quantum circuit simulations (QCS). Designed to take full advantage of the computational power of graphics processing units (GPUs), Q-GPU will result in efficient and scalable QCS. Q-GPU could overcome the current challenges in QCS robustness and efficiency, which has limited the development of quantum computing, to revolutionize the field.
Description
Quantum computing could result in higher speed, higher power computing far outpacing even the best classical supercomputer. In recent years, research into quantum computing has grown attracting billions of dollars in investment. While some success has been observed in research labs, current quantum computing is positioned in the Noisy Intermediate-Scale Quantum (NISQ) era with only limited availability to the public. QCS are required to investigate the constraints and foster quantum computing development including quantum algorithm development and quantum device architecture exploration. QCS are performed on classical computers and as the number of qubits grows the memory costs increase exponentially, resulting in slow simulation and poor scalability. Q-GPU can provide more efficient and scalable QCS.Applications
• Developing algorithms• Validating and evaluating proposed quantum circuits
• Design of quantum machine architecture
Advantages
There are several shortcomings in current QCS approaches. Firstly, discrepancy in supported quantum gates from different QCS software (e.g., Google qsim and IBM Qiskit) which makes algorithm development difficult and non-portable. Secondly, simulation cost increases considerably with increasing numbers of qubits and complexity of quantum circuits, limiting the scalability of any QCS. Thirdly, the execution of a current QCS approaches lacks optimization to fully take advantage of underlying computing resources and limits efficiency.Q-GPU uses modern GPUs to develop a high-performance, scalable, general-purpose QCS. GPUs have rich computational parallelism which is harnessed by Q-GPU to better use the GPU, compared to existing approaches. Zero-valued and non-zero state amplitudes are handled differently and separately, unlike many current approaches which use single data compression on all states. Additionally, data movement between the central processing unit (CPU) and GPU is minimized to improve efficiency.