.png?length=1200&name=ALIRO%20Scaling%20Simulations%20of%20Distributed%20Quantum%20Computing%20(DQC).png)
Scaling Simulations of Distributed Quantum Computing (DQC)
Quantum networks are being built to enable Distributed Quantum Computing and scale-up quantum computation. Simulation of Distributed Quantum Computing is becoming increasingly important for testing, optimization, and feasibility analysis of networking QPUs together. Networking QPUs together requires specialized simulation to accurately model entanglement generation and distribution, dynamic circuit behavior, and increasing sophistication of the networks being simulated across different levels of scale.
The Simulation Landscape: CPUs vs GPUs
The computational resources required to simulate DQC depends on the scale and complexity of the problem. At small scales, operations like basic entanglement distribution can be effectively simulated using local CPUs. These simulations employ traditional state representations like state vectors and density matrices and have minimal computational overhead. As the scale increases, the demand for more advanced computational resources grows too.
As scale of the network increases, the assistance of high performance computing resources becomes more and more necessary. For example, simulating a large DQC network where each of the nodes has a circuit or in cases where qubit size becomes larger. In these cases, it becomes necessary to leverage multi-core CPUs and high performance computing in the cloud, and/or GPUs, which are especially useful for these high dimensional states with large matrix operations.
For large-scale simulations, traditional state representations become impractical. These simulations must use alternative approaches to modeling at this scale, such as stabilizer formalism and tensor networks.
Benchmarking GPU vs. CPU Performance
To assess the performance of GPUs and CPUs in DQC simulations, some benchmark tests have been conducted using an NVIDIA A100 GPU.
The tests fixed the circuit depth but varied the number of qubits from 18 to 27. The results show that CPU performance follows an exponential curve, with runtime increasing dramatically beyond 22 to 24 qubits.
The GPU runtime increased at a much slower rate. This demonstrates a performance advantage when simulating higher qubit counts using GPUs versus CPUs. At peak tested capacity, GPUs provided a tenfold improvement in speed over CPUs.
Trusting Distributed Quantum Computing Simulations
An important consideration in quantum computing research is how trustworthy distributed quantum computing simulations are in matching reality. This is a significant challenge, particularly when simulating systems that do not yet have a hardware implementation. Simulations are often used to explore how a system might behave before it is physically constructed. One way to build up confidence in the accuracy of simulations is by simulating a portion of the system and comparing the results to real-world measurements. Even when large-scale simulations cannot be directly tested against experimental data, confidence in their accuracy can be built by confirming the correctness of each simulated component.
The reliability of quantum simulations depends on an iterative feedback loop between simulation and experimentation. For example, measurements taken in the lab can inform noise models used in the simulations. Simulations are refined with these noise model and can be used to predict outcomes, which can then be tested experimentally. This feedback loop helps researchers converge on models that accurately simulate real quantum systems.
The rapid advancements in quantum computing demand a proactive approach to simulation, ensuring that researchers and engineers can stay ahead of emerging challenges. A robust simulation strategy incorporates the right high performance computing resource and uses different levels of abstraction to inform the design and implementation of Distributed Quantum Computing.
For more details about the utility of quantum network simulators for modeling Distributed Quantum Computing architectures, please see our on-demand webinar, Distributed Quantum Computing (DQC): Technical Implementation Considerations.
