A GPU-based simulator handles circuits with 180,000 qubits

Muhammad Osama, Dimitrios Thanos, and Alfons Laarman of Leiden University have developed QuaSARQ, a GPU-accelerated simulator capable of handling quantum circuits with 180,000 qubits. The new simulator addresses a critical bottleneck in quantum computing research by enabling the modeling of systems previously too large for effective simulation.

QuaSARQ achieves up to a 105x speedup over existing simulators like Stim, Qiskit-Aer, Qibo, Cirq and PennyLane on demanding instances, while also reducing energy consumption by over 80%. These results demonstrate that their parallel algorithms can significantly advance the scalability of stabilizer-circuit simulation, particularly for workloads involving extensive measurements and sampling.

QuaSARQ Simulator Achieves 180,000 Qubit Stabilizer Circuit Simulation

QuaSARQ reformulates core challenges in stabilizer simulation, specifically Gaussian elimination and measurement updates, into GPU-optimized processes that eliminate sequential dependencies and maximize memory efficiency. This approach led to the development of QuaSARQ, a GPU-accelerated simulator designed to handle circuits with a large number of qubits and extensive sampling requirements. Benchmarking across a suite of tests, reaching 180,000 qubits and a depth of 1,000, equivalent to roughly 130 million gates, demonstrated substantial runtime improvements and an over 80% reduction in energy consumption on demanding instances.

The simulator consistently outperformed established tools like Stim, a CPU-optimized stabilizer simulator, alongside Qiskit-Aer, Qibo, Cirq and PennyLane. These gains are not merely incremental; QuaSARQ achieved a speedup of up to 105x compared to these existing simulators. This performance advantage is particularly pronounced in many-shot sampling, a process important for evaluating quantum error correction protocols with large workloads.

The open-source implementation of QuaSARQ is publicly available on GitHub, allowing researchers to build upon and validate the findings. The need for faster simulation stems from the requirements of quantum error correction, a necessary step before quantum computers can reliably execute complex algorithms. Current simulation methods are limited to tens of thousands of qubits, insufficient for testing modern error correction protocols. Researchers identified the sequential nature of standard measurement algorithms as a key obstacle to GPU acceleration, a pattern that inherently limits parallel processing.

“The obstacle is that the standard measurement algorithm is inherently sequential: each update depends on the result of the previous one, which is precisely the pattern a GPU cannot speed up,” the study explains. By redesigning these algorithms, QuaSARQ allows GPUs to simulate increasingly complex quantum systems and advance the field of quantum computing.

GPU-Tailored Primitives Overcome Sequential Measurement Bottlenecks

QuaSARQ achieves over 80% energy reduction on demanding instances, a benefit stemming from redesigned algorithms that address limitations in conventional quantum simulation. The team reformulated Gaussian elimination and measurement updates into GPU-tailored primitives, eliminating sequential dependencies that previously hindered performance. The researchers discovered that this chain of updates could be rewritten as a prefix-XOR scan, an operation ideally suited for parallel processing on GPUs.

Combined with a carefully designed memory layout, QuaSARQ bypasses the bottlenecks that limited prior simulators. The result is a system capable of handling circuits at a scale previously unattainable for many error correction protocols.

QuaSARQ Demonstrates 105x Speedup Versus Existing Simulators

This improvement extends to over 80% energy reduction, addressing a critical constraint in large-scale quantum simulation. This approach bypasses sequential dependencies that traditionally bottleneck stabilizer simulation. The ability to efficiently simulate circuits at this scale is important for designing and operating quantum error correction protocols, which are essential for building reliable quantum computers. QuaSARQ’s ability to model systems previously unattainable marks an advancement.

Energy Reduction and Scalability for Quantum Error Correction Workloads

Beyond simply scaling to larger qubit counts, QuaSARQ demonstrably reduces energy consumption during simulation of demanding quantum error correction workloads, achieving over 80% energy reduction on challenging instances. This efficiency stems from a fundamental reformulation of how measurements, a core component of error correction, are processed, shifting from sequential to parallel operations. The team’s approach uses the inherent parallelism of GPUs to bypass bottlenecks previously limiting simulation scale and speed.

The core of this improvement lies in rewriting the chain of updates inherent in quantum measurement as a prefix-XOR scan, a task ideally suited to GPU architecture. The result is a simulator that not only handles circuits but also does so with a substantial reduction in computational resources. The open-source implementation, available at GitHub, allows other researchers to build upon this work and further advance the field of quantum error correction.

More like this

Quantum SimulationLeibniz Supercomputing Centre pipeline links quantum and classical simulation

Quantum PhysicsPhysicists simulate magnetic monopoles on a quantum spin chain

Quantum PhysicsA hotter quantum system may relax faster, simulations suggest

Quantum AlgorithmsA quantum simulation method avoids extra qubits for observable estimation