Achieving Energetic Superiority Through System-Level Quantum Circuit Simulation (2407.00769v1)

Abstract: Quantum Computational Superiority boasts rapid computation and high energy efficiency. Despite recent advances in classical algorithms aimed at refuting the milestone claim of Google's sycamore, challenges remain in generating uncorrelated samples of random quantum circuits. In this paper, we present a groundbreaking large-scale system technology that leverages optimization on global, node, and device levels to achieve unprecedented scalability for tensor networks. This enables the handling of large-scale tensor networks with memory capacities reaching tens of terabytes, surpassing memory space constraints on a single node. Our techniques enable accommodating large-scale tensor networks with up to tens of terabytes of memory, reaching up to 2304 GPUs with a peak computing power of 561 PFLOPS half-precision. Notably, we have achieved a time-to-solution of 14.22 seconds with energy consumption of 2.39 kWh which achieved fidelity of 0.002 and our most remarkable result is a time-to-solution of 17.18 seconds, with energy consumption of only 0.29 kWh which achieved a XEB of 0.002 after post-processing, outperforming Google's quantum processor Sycamore in both speed and energy efficiency, which recorded 600 seconds and 4.3 kWh, respectively.

• The paper introduces a three-level parallel scheme that distributes tensor network computations across up to 2304 GPUs, achieving a time-to-solution of 17.18 seconds and energy usage of 0.29 kWh.
• The paper employs a hybrid communication strategy with low-precision quantization, reducing inter-node transfer time by nearly 85% while maintaining a fidelity of 0.002.
• The paper challenges quantum supremacy by setting new classical simulation benchmarks, demonstrating that optimized supercomputing can rival quantum processors on complex tasks.

Achieving Energetic Superiority Through System-Level Quantum Circuit Simulation

The paper "Achieving Energetic Superiority Through System-Level Quantum Circuit Simulation", presents a detailed and analytical exploration into the development of large-scale system technology optimized for the simulation of quantum circuits, specifically random quantum circuits (RQCs). This paper directly addresses the milestone set by Google's Sycamore quantum processor, tasked with quantum supremacy through random circuit sampling.

Overview

The primary focus of the paper is the creation of a scalable system leveraging tensor networks for the effective simulation of large quantum circuits. The authors propose a multilayered optimization approach, encompassing global, node, and device levels, to break past prior computational limits. This is achieved by implementing an extensive parallel architecture capable of distributing a simulation's computational burden across up to 2304 GPUs, resulting in peak computational performance of 561 PFLOPS in half-precision.

Key Contributions and Techniques

1. Three-Level Parallel Scheme:

The authors introduce an intricate three-level parallel scheme to maximize computational efficiency by leveraging distributed-memory systems: - Global Level: The original tensor network is split into parallel, independent sub-networks. - Multi-Node Level: Responsibilities are distributed across nodes interconnected via InfiniBand, emphasizing node-level slicing and recomputation strategies. - Device Level: Involves breaking down data further into chunks that are handled by individual GPUs within each node, maximizing intra-node bandwidth utilization via NVLink.

1. Hybrid Communication Strategy: A hybrid communication model is proposed to blend inter-node and intra-node data exchanges, carefully balancing communication load to optimize for both performance and energy efficiency.
2. Low-Precision Quantization: To reduce the overhead of data transfer, particularly inter-node transfers which are naturally more bandwidth-constrained, a low-precision quantization approach is applied. The use of int4 quantization with dynamic group sizes achieves substantial reductions in communication time with minimal fidelity loss, illustrating nearly 85% lower communication time compared to using full precision data.
3. Einsum Extension for Complex-Half Precision: The paper extends the traditional einsum approach to support complex-half precision operations, essential for squeezing more computation within the limited memory space provided by each GPU while maintaining computational accuracy.
4. Special Case Optimizations:
   • Recomputation Techniques: Applied to large intermediate tensors to reduce node requirements and computation redundancies.
   • Sparse State Tensor Contraction: Refinements to tensor multiplication in the sparsely populated regime typical in late-stage network calculations, leveraging high-speed tensor core computations.

Results

The experimental verification showcased performance significantly exceeding Sycamore's benchmarks. Notable results include:

• Achieving a time-to-solution of 17.18 seconds with an energy consumption of only 0.29 kWh for tensor networks sized up to 32TB with post-processing. This outperforms Sycamore's record of 600 seconds and 4.3 kWh.
• An uncompromised fidelity of 0.002 was maintained across simulations, preserving the accuracy required for computational integrity in quantum experiments.
• Scalable efficiency in computational tasks is demonstrated with a linear decrease in time-to-solution relative to the number of GPUs utilized, evidenced by strong scaling characteristics between 128 and 2304 GPUs.

Implications and Future Directions

The work challenges Google's assertion of quantum supremacy by demonstrating that classical simulations, when paired with state-of-the-art hardware and algorithmic advancements, can outperform quantum processors on certain tasks. It sets a new benchmark for classical computational techniques in the domain of quantum circuit simulations, suggesting that the boundary between classical and quantum advantage is more fluid than previously considered.

From a practical standpoint, the proposed methods and results reveal that classical supercomputers still have significant untapped potential in the landscape of computational physics and quantum computing. As quantum hardware continues to evolve, the theoretical and practical implications of this research suggest a thriving competitive space between classic and quantum hardware.

Future directions could explore extending these techniques to more complex quantum systems or other problem domains such as condensed matter physics or combinatorial optimization, potentially driving advancements in numerous computational fields.

This research offers substantial contributions to both theoretical constructs and practical implementations, marking a significant step forward in the domain of large-scale quantum circuit simulations.