Quantum Simulators as a Managed Service

The quantum computing ecosystem is rich, with over 30 open-source quantum simulation libraries, each offering unique features and capabilities. Our team at BlueQubit has undertaken the colossal task of benchmarking these libraries to identify the most efficient tools for large-scale simulations. Our findings, some of which can be found in our benchmarks article, reveal that GPU-based simulators leveraging NVIDIA cuQuantum, a toolkit designed to accelerate computing workflows on GPUs, stand out for their performance in large-scale experiments.

‍

cuQuantum offers plugins for popular quantum computing frameworks such as Qiskit, Pennylane, and QSim, making it a versatile choice for a wide range of applications. BlueQubit integrated this powerful library into its service stack to provide a managed quantum simulation service. Users can simply submit their quantum circuits—created in Qiskit, Cirq, or Pennylane—to BlueQubit servers, and we handle the rest. Our service eliminates the need for users to deal with containerization, virtual machines, GPU driver setup, and other technical complexities, offering a straightforward path to simulating quantum circuits on large GPU clusters. This approach enables up to 36-qubit universal state-vector simulations without any preparatory work required from the users.

‍

Quantum ML and Differentiable Quantum Programming on GPUs

Quantum Machine Learning (QML) is an exciting frontier where quantum computing intersects with machine learning. Pennylane, a popular tool for QML, is notably efficient for differentiable quantum programming, a technique crucial for optimizing quantum circuits. Thanks to the integration of cuQuantum with Pennylane through the lightning.gpu plugin, developers have access to differentiable quantum programming capabilities on GPUs, a feature that significantly enhances the speed and efficiency of QML experiments.

Our team at BlueQubit has developed modifications to the cuQuantum-Pennylane integration, propelling BlueQubit to the forefront of quantum ML research. We have applied these advancements in our paper, “Hierarchical Learning for Quantum ML: Novel Training Technique for Large-Scale Variational Quantum Circuits,” where we demonstrate the training of 27-qubit, 1000-parameter Variational Quantum Circuits (VQC) models on GPUs. This breakthrough was made possible by employing a novel technique that modifies adjoint differentiation, allowing the computation of gradients for all parameters in the VQC model with only two circuit passes. This method significantly reduces the computational overhead typically associated with training large-scale quantum circuits, marking a milestone in the field of QML.

Multi-Node GPU Simulation: The Sky’s the Limit

One of the most significant advancements in quantum simulation is the ability to perform multi-node GPU simulations. BlueQubit has harnessed this capability to transcend the traditional limits of quantum computing simulations. Traditionally, simulating more than 36 qubits was a formidable challenge due to the exponential increase in computational resources required. However, with NVIDIA cuQuantum’s support for multi-node GPU simulation, BlueQubit users are equipped to go beyond these limitations with only a few lines of code.

The number of qubits we can simulate is now primarily limited by the number of GPUs we can harness in unison. This breakthrough has made it possible to simulate quantum systems that were previously thought to be out of reach, opening new horizons for quantum research and development. This powerful capability is currently offered exclusively in the Team tier of our services and is an invaluable tool for advanced research teams seeking to push the boundaries of quantum computing

Quantum-Assisted Training of Boltzmann Machines on GPUs

Our R&D team at BlueQubit is also using fast GPU simulators to perform quantum-assisted training of Boltzmann Machines (BMs) on GPUs as part of our efforts to find quantum advantage. Quantum-assisted training of BMs represents a fascinating intersection of quantum computing and machine learning, offering the potential for faster convergence in certain sampling algorithms compared to their classical counterparts. The speed and performance on GPU simulators is crucial for our iterative process, allowing us to rapidly test and refine our quantum algorithms.

Our preliminary findings suggest that quantum-assisted training of BMs, when simulated on GPUs, show promising results in terms of time to convergence for certain sampling problems.  As we continue to delve into this exciting area, we are preparing a paper that will provide more technical details on our methodology, findings, and the implications of our research. This forthcoming publication aims to contribute to the broader understanding of quantum computing’s potential, particularly in the realm of machine learning and optimization. 

Tensor Network Simulators: The New Approach to Quantum Programming

Another innovative feature that BlueQubit offers is the use of Tensor Network simulators. Tensor Networks provide a novel approach to simulating certain types of quantum circuits that are low in entanglement, allowing us to overcome the traditional memory limitations of 36-40 qubits. With this technique, we can simulate circuits with 100+ qubits in some cases. This is a significant leap forward, as it enables the simulation of much larger and more complex quantum systems than ever before.

The ability to simulate such large systems is crucial for advancing our understanding of quantum mechanics and for developing new quantum algorithms. This feature, like multi-node GPU simulation, is also experimental and currently exclusive to the Team tier. It represents another step toward making quantum computing more accessible and practical for a broader range of applications.

The Road Ahead

Ten years ago, the quantum processing units (QPUs) available could be simulated on a laptop. Today, we have quantum computers with hundreds of qubits, far beyond the capability of any personal computing device. To run large-scale experiments involving circuits run on actual quantum computers, researchers now require substantial computational resources, including large clusters of CPUs and GPUs.

In this new era of quantum computing, it’s crucial to have robust benchmarks and baselines for large-scale numerical experiments, particularly in the QMLsphere. This is where BlueQubit excels. BlueQubit uses NVIDIA’s GPU emulators to help its users run large-scale quantum simulations on GPU clusters, offering an unprecedented level of simulation capability.

At BlueQubit, we are committed to democratizing access to quantum computing resources, enabling researchers, developers, and businesses to explore quantum algorithms and their viability on today’s quantum hardware without the barriers traditionally associated with high-performance quantum simulations. Through our managed quantum simulator service, powered by NVIDIA GPUs, we are opening new avenues for innovation and exploration in the quantum computing domain, making the quantum future more accessible than ever before.