CUDA-Q Realtime
Type: Technology Tags: NVIDIA, CUDA-Q, cudaq-realtime, NVQLink, realtime, quantum control, QEC, GPU-QPU Related: CUDA-Q, CUDA-QX, NVIDIA-NVQLink, NVIDIA-Quantum, Ising-Decoding, NVIDIA-BlueField-DPU, NVIDIA-ConnectX-InfiniBand, NVIDIA-Holoscan Sources: https://nvidia.github.io/cuda-quantum/latest/using/realtime.html, https://nvidia.github.io/cuda-quantum/blogs/blog/2026/03/16/launching-cudaq-realtime/, https://www.nvidia.com/en-us/solutions/quantum-computing/nvqlink/ Last Updated: 2026-04-29
Summary
CUDA-Q Realtime, also referenced as cudaq-realtime, is the CUDA-Q library for tightly coupling GPU-accelerated compute to a quantum processor control system through a low-latency networking layer. It is the software layer that makes NVIDIA-NVQLink programmable for microsecond-class GPU-to-quantum-controller feedback loops.
Detail
Purpose
Fault-tolerant quantum computing requires realtime classical processing while the QPU is operating. CUDA-Q Realtime gives developers and system integrators an API and networking layer for callbacks between GPUs and quantum controllers, enabling QEC decoding, calibration, and other realtime control workflows.
Platform role
- Requires a host system with an NVIDIA GPU and ConnectX-7 or BlueField NIC, plus an FPGA connected to the NIC.
- Provides realtime co-processing between FPGA and CPU-GPU systems.
- Provides the low-latency networking layer of the NVQLink architecture.
- Uses an Ethernet/RoCE data path and references Holoscan Sensor Bridge for high-bandwidth transfer.
- Became publicly available through CUDA-Q around the GTC 2026 NVQLink release window.
NVIDIA context
Use this page for the software/API layer. Use NVIDIA-NVQLink for the architecture and hardware integration story, CUDA-QX for QEC and solver libraries, and CUDA-Q for the broader programming model.
Connections
- CUDA-Q - CUDA-Q Realtime is part of the CUDA-Q platform.
- NVIDIA-NVQLink - realtime library provides the programmable low-latency NVQLink path.
- CUDA-QX - CUDA-Q QEC uses realtime decoding paths for active error correction.
- Ising-Decoding - NVIDIA predecoder models are used in realtime QEC examples and latency-sensitive decoding workflows.
- NVIDIA-BlueField-DPU and NVIDIA-ConnectX-InfiniBand - NVIDIA networking components used in realtime host designs.
- NVIDIA-Holoscan - Holoscan Sensor Bridge is referenced as part of the high-bandwidth data-transfer layer.
Source Excerpts
- NVIDIA CUDA-Q docs describe CUDA-Q Realtime as a library for coupling GPU-accelerated compute to a quantum processor control system through networking, providing the low-latency networking layer of NVQLink.
- NVIDIA’s CUDA-Q Realtime launch blog positions the library as the API for microsecond-latency callbacks between GPUs and quantum controllers.