ORNL Quantum Computing Blueprint: Quantum-HPC Integration
ORNL Quantum Computing
Oak Ridge National Laboratory gives the design for seamless quantum-HPC integration.
The U.S. Department of Energy's Oak Ridge National Laboratory (ORNL) has released a complete software stack design to seamlessly merge HPC with QC. In an ORNL publication, this new study proposes a hardware-agnostic framework to overcome major technological barriers to combining these two computational paradigms, providing a foundation for future scientific discovery.
As GPUs improve traditional computing through heterogeneous integration, quantum computing is expected to operate as a specialised accelerator for activities that suit its capabilities rather than replace it. QPUs are meant to speed up quantum-advantaged algorithms in scientific procedures, like GPUs did for ORNL's Frontier supercomputer, which broke the exascale barrier. A solid integration plan with standard HPC systems is needed.
This study improves on previous ORNL quantum computing work by highlighting software architecture and design practicalities and providing more implementation guidelines. The proposed architecture's primary innovations include:
One resource management system for quantum and classical resources. A flexible Quantum Programming Interface (QPI) that abstracts hardware-specific details for application developers. The Quantum Platform Manager (QPM) API simplifies quantum hardware integration. A complete set of quantum circuit optimisation and implementation tools. A quantum gateway interface for several quantum hardware systems that decreases workload and oversubscription. The framework supports noisy intermediate-scale quantum (NISQ) devices and fault-tolerant quantum computers while being hardware-agnostic and compatible with HPC operations.
GPUs' successful integration into traditional computing inspired the design. Quantum applications require compilers that pre-process code, partition it into host and quantum segments, and optimise quantum operations into an intermediate representation (IR), just like GPU programming. This quantum-enabled compilation cycle allows hardware-specific optimisations like GPU-like Just-In-Time (JIT) compilation during runtime.
The QC/HPC software stack relies on efficient administration to balance application productivity and resource optimisation. The study defines three application patterns: nearly equal utilisation, low quantum/high classical, and high quantum/low classical based on quantum and classical resource needs. Due to its scarcity, quantum gear must be allocated more efficiently than standard HPC equipment.
Interleaved allocations allow discrete reservations that may overlap or form a linked series, whereas simultaneous allocations allow quantum and classical computing resources to be booked for the same time. ORNL is testing SLURM's heterogeneous job (hetjob) functionality to manage HPC and quantum resources simultaneously. A sbatch script may request one quantum QC node and 10 HPC nodes. Credit systems are intended to provide “soft allocations” and quality of service guarantees for quantum workloads to prevent HPC resources from lying idle while quantum results are made.
The suggested software stack has discrete levels for suitable abstraction. QPI APIs handle initialisation, resource management, execution control, tool setup, device management, and result processing. The hardware abstraction layer Quantum Platform Manager (QPM) streamlines quantum job submission, results retrieval, and device status queries. The QPM's plugin design lets quantum hardware suppliers write plugins and share scheduling and communication capabilities.
The Quantum Toolchain API, which formalises quantum circuit transformation tools, is crucial. These tools polish QIR or OpenQASM quantum programs for hardware compatibility by optimising gate reduction and circuit cutting. These technologies use HPC resources to make computationally intensive modifications.
Using a NWQ-Sim backend, a variational quantum linear solver (VQLS) hybrid quantum-classical application evaluated the design. This hands-on experiment showed the framework's ability to manage complex hybrid workflows and revealed application inefficiencies such optimiser selection's impact on circuit inspections. The framework also helps programs plan for parallel execution by identifying sequential circuit construction delay issues.
The software stack integrates easily with workflow orchestration frameworks like Pilot-Quantum to ensure tight integration. OLCF's Secure Scientific Service Mesh (S3M) provides a foundation for secure access and integration of quantum resources in the HPC environment while maintaining operational security and trust boundaries. Telemetry collects quantum and classical operational parameters for real-time monitoring and historical analysis to optimise performance.
Given that quantum computers will remain constrained, the framework provides a flexible and scalable modelling environment. This environment uses a QPM plugin to integrate simulator backends like TNQVM and NWQ-Sim, allowing researchers to test and debug applications on HPC nodes before deploying them on quantum hardware.
This ORNL effort provides a solid foundation for integrating quantum computing into HPC facilities. It addresses latency, resource management, and workflow optimisation, paving the way for quantum computing to speed up computational and scientific processes and create new research opportunities.
Combining modelling and simulation, artificial intelligence, and quantum computing into powerful, adaptive tools to accelerate scientific discovery is a major problem for computational research. This plan is essential to the future.
