The Signal
Qblox Reports Sub-4-Microsecond GPU–Quantum-Control Latency
Qblox says its Cluster control system reached round-trip latency below 4.0 microseconds with an NVIDIA RTX PRO 6000 Blackwell GPU and below 7.2 microseconds with DGX Spark. The October 22, 2026 announcement describes a GPU-linked control se
Quantum.world Team · Written with AI · fact-checked ·
Below 4 microseconds on the Blackwell system
Qblox says its quantum-control system reached round-trip latency below 4.0 microseconds when connected to an NVIDIA RTX PRO 6000 Blackwell GPU, and below 7.2 microseconds with NVIDIA’s DGX Spark. Announced October 22, 2026, the results put GPU computing inside the control loop that links quantum-system hardware with classical processing.
The company describes integrating its Qblox Cluster with NVIDIA GPUs through NVQLink, NVIDIA’s architecture for connecting quantum-system controllers to accelerated computing. The release also says a single DGX Spark desktop can host quantum, artificial-intelligence and simulation workloads through CUDA-Q, with the same code intended to carry over to GPU servers in data centers.
Qblox CEO and co-founder Niels Bultink said: “Quantum computers will scale only as fast as the classical control systems around them.” He said the integration makes GPUs part of that control loop, giving quantum systems access to computing resources in real time.
What the latency figure measures—and what it does not
The sub-4-microsecond result is attached to the RTX PRO 6000 Blackwell configuration; the DGX Spark result is below 7.2 microseconds. NVIDIA describes NVQLink’s latency benchmark as a round-trip callback between a quantum system controller and a host, exercising communication through the controller-host link. That is a link-and-control measurement, not, by itself, a reported computation on a quantum processor.
Qblox’s announcement does not name a processor, report a qubit count, or provide the benchmark’s detailed test conditions. It also does not specify how much of the measured time comes from communication versus any processing performed during the loop. The figures therefore establish a measured connection performance claim, not a demonstrated end-to-end quantum-error-correction result.
Calibration and decoding are the intended workloads
Qblox says its Q1 sequence processors and LINQ architecture connect the Cluster to the GPU with the latency and throughput needed for the integration. GPU-accelerated calibration routines can update qubit parameters while experiments are running, rather than waiting until an experiment pauses, according to the release. The company also says syndrome data can reach GPU-based decoders for error processing.
Those tasks place different demands on the connection: calibration adjusts operating parameters, while error correction requires measurement results to be decoded quickly enough to inform a response. A low-latency link can support both, but the timing figure alone does not establish decoder speed, correction accuracy or performance on a live device.
From NVQLink plans to measured hardware
Qblox frames the announcement as a step beyond its earlier NVQLink work, announced in October 2025, and its March 2026 support for the CUDA-Q real-time API. The March announcement described an FPGA-to-GPU feedback architecture with Oxford Quantum Circuits; the new release reports specific latency measurements for the Cluster paired with NVIDIA GPU systems.
For teams building hybrid systems, the practical question is whether the reported link timing holds under the full workload: real device readout, GPU-side calibration or decoding, and a return signal that affects subsequent control. Qblox has published the headline measurements, but its announcement does not provide those operating details or identify a quantum processor used in the test.