Qblox links NVIDIA DGX Spark to quantum control with sub-7.2 µs latency
Qblox says it has integrated NVIDIA DGX Spark with its quantum controller via NVQLink, targeting real-time calibration and error correction.
Qblox says it has connected NVIDIA’s DGX Spark desktop AI system to its quantum system controller using NVIDIA NVQLink, reporting measured round-trip latency below 7.2 microseconds on DGX Spark and below 4.0 microseconds on an NVIDIA RTX PRO 6000 Blackwell. According to the company announcement, the goal is to put GPU computing directly into the real-time quantum control loop for workloads such as calibration and quantum error correction.
What was announced
The announcement, published October 22, marks a further step in Qblox’s work with NVIDIA around hybrid quantum-classical computing. Qblox says the new setup combines its control stack with NVIDIA DGX Spark so that a single desktop machine can host quantum, AI and simulation workloads through CUDA-Q.
According to the announcement, developers can build applications on desktop hardware and then move the same code to GPU servers as systems scale. NVIDIA describes CUDA-Q as an open quantum computing platform spanning CPU, GPU and QPU resources in one program.
Qblox CEO and co-founder Niels Bultink said classical control will be a limiting factor in scaling quantum computers, adding that the integration is intended to make GPUs part of that control path.
The technical details
Qblox attributes the low-latency link to its Q1 sequence processors and LINQ interconnect, which it says provide the throughput needed for its Cluster control system to work alongside GPUs. In this architecture, GPU-accelerated routines can update qubit parameters while experiments are running rather than only between runs.
The company also says syndrome data for quantum error correction can be sent to GPU-based decoders with sufficiently low latency to support real-time identification and correction of errors. That is a notable distinction: in many hybrid workflows, classical post-processing sits outside the critical timing path, while here the stated aim is direct participation in live control.
The reported measurements are hardware-specific. Qblox says round-trip latency came in below 4.0 microseconds on the RTX PRO 6000 Blackwell and below 7.2 microseconds on DGX Spark. The announcement does not provide a fuller benchmarking methodology or independent validation.
Why it matters
For quantum processors, fast classical feedback is increasingly important as systems move toward more complex calibration and error-correction routines. If GPUs can be inserted into the control loop without adding too much delay, they could handle computationally heavy tasks that would be difficult to execute in fixed-function control electronics alone.
That is particularly relevant for decoder workloads in error correction and for adaptive calibration, where decisions must be made quickly enough to affect the ongoing experiment. Qblox’s announcement suggests an effort to bridge lab-scale development and larger-scale deployment by using the same CUDA-Q software path on both desktops and data center GPU systems.
What to watch
This latest step builds on earlier milestones cited by Qblox: previous work with NVQLink announced in October 2025 and support for the CUDA-Q realtime API in March 2026. The new claim is not about a future architecture, but about measured performance on hardware now integrated with the company’s controller.
The next questions are practical ones. Researchers and customers will want to see how the latency figures translate into specific applications, how broadly the approach fits different qubit modalities, and whether the end-to-end system can sustain these timings under larger experimental loads. For now, the announcement points to a clearer role for GPUs inside quantum control, not only beside it.
Sources
Facts on this page come from the sources below, retrieved by DesignQuantum.
- Qblox Integrates NVIDIA DGX Spark With Quantum System Controller thequantuminsider.com
- NVIDIA CUDA-Q developer.nvidia.com