Ilustración en 3D de un refrigerador de dilución con sus platos dorados y los cables planos que llevan señales a los procesadores cuánticos

· Translated from the Spanish original

Quantum computing

Qblox links its quantum controller to an NVIDIA DGX Spark with 7.2 microseconds of latency

The Dutch company measured under 4.0 µs with a server GPU and under 7.2 µs with a desktop machine. It is measured latency, not a quantum result.

What happened

  • Qblox reported on October 22, 2026 that it integrated an NVIDIA DGX Spark with its quantum system controller using NVIDIA NVQLink.
  • It measured round-trip latency under 4.0 microseconds with an NVIDIA RTX PRO 6000 Blackwell GPU and under 7.2 microseconds with the DGX Spark.
  • With NVIDIA CUDA-Q, a single desktop machine can host quantum, AI and simulation workloads. The same code, the company says, then moves to GPU servers.
  • The GPU enters the real-time control loop: it calibrates qubit parameters while the experiment runs and decodes syndrome data to correct errors.

Why it matters

  • The announcement moves the discussion from architecture to measurement. In October 2025 Qblox presented its work with NVQLink; now it delivers hardware numbers.
  • The 7.2 µs matters because error correction requires deciding faster than qubits degrade. A decoder that arrives late corrects nothing.
  • For a Chilean university setting up a quantum lab, the consequence is cost: testing the classical–quantum loop on a desktop machine avoids buying a server before knowing whether the design works.

The number

7.2 µs. Round-trip latency measured with the DGX Spark; with the Blackwell server GPU, under 4.0 µs.

Context

Quantum error correction is the bottleneck to reaching useful machines. On October 19 Quantinuum reported 99.9995% logical fidelity on its Helios system; Qblox attacks another piece, the electronics and classical computing around the qubits. The release builds on March 2026 work with the CUDA-Q real-time API.

What’s next

  • No dates for an error-correction demonstration with this setup; the release reports no results with logical qubits.
  • Scaling up to GPU servers is described as a path, with no timelines.

Bottom line

The quantum race is also played off the chip: Quantinuum reached 5 errors per million operations on a logical qubit, and Xanadu secured up to US$10 million from DARPA for its photonic route. In Germany, Bechtle will give an aerospace center access to an IQM quantum cloud.

Sources

Edited by Rodrigo Cornejo. How we select and verify: who writes these notes.

Related notes

← All notes