loading
loading
COMPUTE Nvidia Simulates Its Way Into Google's Magic State Cultivation Party There’s a lot of innovation happening in the quantum computing field as it moves closer to the promised land of practical, scalable fault-tolerant systems. That said, reducing errors in calculations remains a core challenge that needs to be solved. Qubits are extremely fragile, can easily fall apart and lose their coherence, and a quantum system that makes too many errors is not useful. Quantum computing vendors create logical qubits from a collection of many physical qubits that can detect, isolate, and fix errors without interrupting the underlying quantum state. But it’s difficult to scale quantum systems when a single logical qubit could require thousands of physical qubits. And the scalability difficulties don't end there. Universal quantum systems – those that can run all operations already performed by classical servers plus those specific to quantum computers – need to include both Clifford gates (like S and CNOT gates) and non-Clifford gates (like T gates) to perform quantum algorithms. Most error correction codes can handle Clifford states, but not non-Clifford gates. In 2004, researchers introduced magic gates, which are prepared by the computer and injected into the computation. They are key to universal quantum computation, but they're also fragile, difficult to create with high fidelity, and can't be used again, so more have to be continuously created. Magic state distillation was used to offset this by collecting many noisy – or inaccurate – states and turning them into fewer, high-fidelity magic states. However, doing all of this takes time and can consume as much as 95 percent of a quantum system's capacity, creating a troubling resource bottleneck. Google researchers in 2024 introduced another technique – magic state cultivation – to more cheaply and efficiently generate high-fidelity magic states. Rather than create huge numbers of noisy states and then discard a lot of them, with cultivation, engineers can put a noisy state into a small area on the quantum chip and then – like a filter – run measurements via special coding techniques to find errors and remove flawed tries, eventually growing a lower-error magic state and significantly driving down the hardware overhead needed to run quantum algorithms. Putting Cultivation to Test On October 15, both Google and Nvidia announced projects they say significantly advance the development of magic state cultivation, with Google Quantum AI researchers reporting the first hardware implementation of the method. Running it on Google’s Willow superconducting quantum chip, the demonstration resulted in a 40-fold reduction in errors compared to uncultivated states and created T-states that reached 99.99 percent in logical fidelity. Less than a week later, Nvidia researchers announced that they’d run a 104-qubit simulation on Jupiter – Europe’s first exascale computer housed at the Jülich Supercomputing Centre in Germany and powered by Nvidia GH200 superchips – that tested magic state cultivation, which they said uses 90 percent fewer qubits than other methods. Sam Stanwyck, director of quantum product at Nvidia, told reporters the work of both vendors was a critical step in eventually reaching fault-tolerant quantum computing, and that Nvidia shows the critical role that such simulations can play. “This is complementary to Google's work,” Stanwyck said. “What Google did is a small-scale demonstration of this in hardware. It was a proof-of-concept that magic state cultivation, the approach, actually works on real hardware. The next steps for Google and for everyone are now, how do you design around that and how do you scale it up? What everyone will want to do to enable that is simulate magic state cultivation to understand it much better and design the next generation of processors. What our result does is point the way with tools and a platform for how to do that.” After two years of working to move cultivation from theory to reality, Google scientists outlined the work and results in a paper published in Science. They noted they need to demonstrate the technique at a larger scale to make it more realistic for a practical quantum system, but added that “this achievement takes a key theoretical idea and turns it into an experimentally demonstrated reality.” Nvidia’s Stanwyck said the company’s work illustrated how simulations can push this along, arguing that “in order to be able to effectively design hardware and software around these approaches, you need to simulate it as deeply as possible to understand how it behaves and, especially, how it handles errors.” The vendor – which wants to be the accelerator for future hybrid quantum-classical environments – is not the only one using a digital twin to run simulations to address error correction. Earlier this year, scientists with Amazon Web Services and Quantum Elements – along with the University of Southern California and Harvard University – wrote about their work in the area. Nvidia scientists used its CUDA-Q quantum-classical platform and cuQuantum software development kit to create detailed simulations of hardware errors – including coherent errors – and leakage. “These are the small, small systematic mistakes in quantum operations that accumulate over a computation and limit what the computer can do,” Stanwyck said. “They were able to simulate that in detail and simulate the effect on the algorithm. Now we're now able to test how magic state cultivation responds to these effects, which helps the researchers work out what the quantum hardware needs to do, how to build it, what code to run, and what to improve before building the next version.” Nvidia researchers started the project on a GB200 NVL72-powered system at its Accelerated Quantum Research Center in Boston to develop and validate methods needed to prepare the software to run on a system the size of Jupiter, he said. The development platform was used to scale tests from four GPUs to hundreds, eventually moving it to 8,192 GPUs on Jupiter. “With magic state cultivation, we were able to simulate from the physics to the application and actually let researchers and those who are developing quantum computers connect a deep understanding of errors in the physical qubits and how they affect application runtime so they can design better [magic] state cultivation, design better quantum computers, and ultimately, this is how simulation can help pull in the timeline to useful quantum computing,” he said.
document fetched 2026-10-27 02:54 · canonical https://nextplatform.com/compute/2026/10/26/nvidia-simulates-its-way-into-googles-magic-state-cultivation-party/5305405