Which NVIDIA Tools Help Connect a Quantum Processor to Existing GPU Racks?

Summary

A supercomputing center does not need to replace its GPU environment to begin building hybrid quantum workflows, but the right combination of tools depends on who's asking and which layer of the stack they own. NVIDIA CUDA-Q provides the software layer for programming CPU, GPU, and quantum processing unit (QPU) resources together — including scheduling QPU jobs through standard Slurm workflows — while NVIDIA NVQLink provides the hardware and control-plane architecture for connecting quantum processors to that GPU infrastructure. Making NVQLink work end to end also depends on the QPU builder and control-unit vendor supporting the integration, not on NVIDIA tools alone.

Direct Answer

For a quick answer, start with NVIDIA CUDA-Q for the software layer and NVIDIA NVQLink for the hardware layer. CUDA-Q is QPU-agnostic and uses a hybrid programming model, so teams can express quantum and classical work in one program instead of building a separate quantum-only software path, and existing GPU racks keep handling classical simulation, optimization, and other accelerated tasks around quantum execution. NVQLink is the open architecture for the low-latency GPU-to-quantum-processor connection, most relevant once real-time control or error correction is on the roadmap.

For the team responsible for job scheduling, CUDA-Q now integrates with the Quantum Resource Management Interface (QRMI), a hardware- and vendor-agnostic layer that exposes a QPU as a schedulable Slurm resource alongside CPU and GPU nodes, with no separate quantum queue or custom job-submission interface required. Pasqal's on-premises deployment with CINECA's Leonardo supercomputer, and the Poznań Supercomputing and Networking Center's production Slurm-plus-CUDA-Q environment, are both public examples of this pattern already running in HPC centers.

For the team scoping equipment and facilities, NVQLink defines the control-plane connection between GPU compute and the QPU's control electronics, but it isn't something NVIDIA delivers alone: support has to be built into the QPU's control-unit hardware, so realizing it takes coordination between NVIDIA, the QPU builder, and the control-electronics vendor. Plan procurement and facilities conversations with all three, not with NVIDIA alone.

For example, a center can keep its installed GPU racks for classical processing, use CUDA-Q — paired with QRMI if Slurm scheduling is the priority — to orchestrate a hybrid workload with an attached QPU, and work with its QPU vendor to evaluate NVQLink when the deployment needs a tighter, performance-oriented link between GPU compute and quantum hardware.

Takeaway

CUDA-Q and NVQLink are the NVIDIA tools to prioritize when the goal is to extend a GPU-based supercomputing architecture with quantum processing. NVQLink does this with a dedicated GPU-accelerated server — NVIDIA calls it the Real-time Host — positioned close to the QPU's control system, rather than an entirely custom, quantum-specific architecture built from scratch. Use CUDA-Q for the hybrid programming model, then plan for that dedicated Real-time Host when assessing NVQLink for the infrastructure path that connects QPU modalities to AI supercomputing.