On November 16, 2026, Altera announced a reference design for a quantum system controller that connects an OQC (Oxford Quantum Circuits) quantum processor to NVIDIA GPUs. The configuration controls the quantum processor through OQC's RF signal-conditioning circuitry and works with the GPU side via NVIDIA's NVQLink. The reference design is scheduled for release in January 2027, and Altera is inviting visitors to its booth at SC26, the international supercomputing conference in Chicago, for a demonstration of the full system.
At the center of the system is an FPGA, a semiconductor whose internal circuitry can be reconfigured. It sends control signals to the qubits and reads out measurement results, while offloading computationally heavy processing to the GPU, making the quantum computer's control hardware more efficient.
How Do the FPGA and GPU Divide the Work?
Running a quantum processor (QPU) requires hardware that sends control signals to the qubits and reads out their responses. In this reference design, Altera's Agilex 9 Direct RF FPGA controls OQC's new-generation QPU through OQC-built RF signal-conditioning circuitry. To read out qubit states, the system uses resonators, which resonate with signals at specific frequencies.
The Agilex 9 Direct RF FPGA has built-in analog-to-digital converters (ADCs), which turn analog signals into digital data, and digital-to-analog converters (DACs), which turn digital data into analog signals. According to Altera, this allows qubit control and readout without external RF frequency converters. Frequency multiplexing also lets multiple readout channels share a single RF port, reducing the hardware required.
However, OQC's RF signal-conditioning circuitry sits between the quantum processor and the FPGA, so not all RF-related circuitry is integrated into the FPGA. On the FPGA side, the system handles synchronized qubit control, resonator readout and spectroscopy. These tasks run under synchronized scheduling, controlling the qubits in real time. The GPU, meanwhile, handles calibration and machine-learning-based classification of readout signals. Spectroscopy examines the frequency response of qubits and resonators, and GPU-based calibration adjusts control parameters.
Based on Altera's announcement and NVIDIA's official NVQLink overview, the roles of each component and connection path are summarized below.
| Component / connection path | Processing it handles | What the announcement describes |
|---|---|---|
| OQC quantum processor and RF signal-conditioning circuitry | Conditions RF signals between the QPU and the FPGA | Connected through circuitry developed by OQC |
| Agilex 9 Direct RF FPGA | Real-time control, resonator readout, spectroscopy | Built-in ADCs and DACs eliminate external RF frequency converters; frequency multiplexing shares RF ports |
| NVQLink and GPU host | Connects the FPGA controller to the GPU; runs calibration and readout-signal classification on the GPU | Drives FPGA logic for the quantum processor from CUDA functions |
| Built-in emulation feature | Controller bring-up and algorithm testing | A development feature that allows operation to be verified before physical qubit hardware is available |
Altera says the features to be released in January 2027 have already been validated. The capabilities provided with the software stack include qubit and resonator spectroscopy, qubit initialization, and resonator readout and classification. Altera will also supply implementation examples, including a machine-learning-based classifier that runs on the GPU and an initialization routine.
Consolidating Functions in the FPGA to Shrink and Cut Power in Control Hardware
The problem Altera aims to address with this design is that control hardware grows larger as the number of qubits increases. More qubits require more circuitry to generate control signals and read out measurement results. By consolidating signal converters and control functions in the Agilex 9 Direct RF FPGA and sharing RF ports through frequency multiplexing, Altera aims to make the equipment smaller and more power-efficient. Besides lower latency, the company also lists limiting the additional RF hardware needed as qubit counts grow among its goals.
The reference design and software stack due for release will serve as an implementation example for quantum computer developers building their own control systems. Sam Stanwyck, director of quantum products at NVIDIA, referred to advances in AI and said, "NVIDIA NVQLink gives quantum computer developers an open architecture to put those advances to practical use." The remark reflects the expectation that AI will be built into quantum computer control, for example through GPU-based calibration and readout-signal classification.
How Should NVQLink Connection Performance Be Evaluated?
Evaluating how well a quantum processor and GPU work together requires more than confirming they can connect; it requires measuring the time for control signals and processing results to make a round trip. NVIDIA describes NVQLink in terms of four elements: a GPU-equipped host, a quantum system controller, the network linking them, and the cudaq-realtime API. In compatibility testing, the controller calls the GPU host and the round-trip latency until the response returns is measured.
Required response times vary with the type of QPU and the purpose of the external processing, ranging from milliseconds to microseconds. NVIDIA says it intends to present reproducible network-latency measurements using a common open-source benchmark. The aim is to allow different controllers and connection configurations to be compared under common conditions.
The announcement introduces the reference design's configuration and features, and no specific performance measurements have been published. The Quantum Insider has also published the announcement as a press release, but no independent third-party replication or peer-reviewed paper has been identified. Altera has not disclosed the number of qubits used in the demonstration or the number of trials, and has not provided measured values for readout-signal classification accuracy, round-trip latency or power consumption.
Once the reference design is released in January 2027, developers will be able to examine the control features and software provided and consider applying them to their own quantum systems. Assessing the actual benefits in size and power reduction, and the latency introduced by GPU integration, will require measurement data that specifies the QPU, GPU and network configuration used.
