NVIDIA ships Sandia's logical-qubit benchmark inside its own CUDA-Q toolchain

The layer makes the parts of a resource estimator interchangeable. NVIDIA states the problem it is built against directly: a change to an algorithm, error-correction code, hardware architecture or other QPU component can significantly change the resources needed to run any given application ^[2]. NVIDIA says CUDA-Q Logical lets a researcher design and orchestrate those components and switch between options to find configurations that perform well with logical qubits ^[3].

The headline result in the release comes out of a model. Iceberg Quantum modeled its fault-tolerant architecture for Diraq's qubits and showed that 1,000 logical qubits can be created with 150,000 physical qubits, roughly 10x fewer than Diraq's previous estimates ^[9]. That is 150 physical qubits per logical qubit ^[13]. Run the 10x backwards and the earlier estimate sat near 1,500 physical per logical, about 1.5 million physical qubits for the same thousand ^[14]. NVIDIA's announcement leaves out the physical error rates assumed in that model, and it gives no QUOPS score for any machine ^[17]. The 150,000 figure moves a hardware roadmap only if fabricated Diraq devices meet the error rates Iceberg's architecture assumed.

What Fermilab measured was a workflow. Researchers there used CUDA-Q Logical to validate prior results and evaluate physical qubits, runtimes and other resource requirements across different error-correction approaches and quantum hardware ^[7], and NVIDIA puts the compression at five months to three weeks, a 7x speedup ^[6]. Five months is about 21.7 weeks, and 21.7 divided by three is 7.2 ^[15]. The multiplier is the ratio of two project durations, and the longer one was spent on tooling. "Using CUDA-Q Logical, our team explored combinations of these resources in just three weeks, compared with what would have typically taken about five months of building specialized infrastructure," said Anna Grassellino, chief technology officer at Fermilab and director of the Superconducting Quantum Materials and Systems Center ^[8].

QUOPS was developed by Sandia National Laboratories as an independent cross-platform benchmark of the progress quantum systems are making toward utility-scale applications, and it is now available in CUDA-Q ^[4]. NVIDIA describes it as a hardware-agnostic, open method to benchmark progress toward shared goals for fault-tolerant machines ^[12]. Sandia's authorship makes the benchmark independent of NVIDIA's hardware; the distribution channel is NVIDIA's software. NVIDIA controls that channel, so the number a QPU maker quotes comes out of NVIDIA's toolchain. A cross-vendor comparison holds only if every vendor's device is modeled through the same code path. "Quantum computing is maturing into an era of logical qubits, and researchers need an open, customizable platform capable of representing all aspects of a fault-tolerant system," said Timothy Costa, vice president and general manager of quantum at NVIDIA ^[10].

NVIDIA says progress in quantum computing has historically been measured through advances in physical qubits, including rising qubit counts and improving fidelity ^[11]. The release lists Infleqtion, IQM Quantum Computers, Fermilab and Sandia among the QPU makers and labs already using CUDA-Q Logical ^[5].