英伟达利用德国于利希超算中心的“朱庇特”超级计算机，模拟了一台能够自我纠错的104量子位量子计算机。该公司表示，这是迄今为止对此类机器进行的最大规模全系统模拟。
Nvidia has simulated a 104-qubit quantum computer that can correct its own errors, using JUPITER, the supercomputer at Germany’s Jülich Supercomputing Centre. The company said it is the largest full simulation of such a machine so far.
量子位是量子计算机的基本构建模块，但它们极其敏感，极易受到干扰。即使来自控制设备的微量噪声也会导致错误。
Qubits are the building blocks of quantum computers, but they’re extremely sensitive to interference. Even small amounts of noise from the control equipment can cause errors.
容错量子计算机能够在运行过程中检测并纠正这些错误。大多数专家认为，这是让量子计算机走向实用的关键一步。
A fault-tolerant quantum computer can detect and correct those errors while running. Most experts consider this a key step towards making quantum computers practically useful.
研究团队测试了一种名为“魔法态培育”的技术，该技术有助于量子计算机执行特定计算。这些特殊的量子态是许多运算所必需的，但创建它们可能需要大量额外的量子位。英伟达表示，其方法所需的量子位比早期方法减少了90%。
The team tested a technique called magic state cultivation, which helps quantum computers carry out certain calculations. These special quantum states are needed for many operations, but creating them can require many extra qubits. Nvidia says its method uses 90% fewer qubits than earlier approaches.
此次模拟还考虑了影响真实量子硬件的噪声。这建立在于利希研究人员的工作基础之上，他们曾于2025年11月在“朱庇特”上、在理想的无噪声条件下模拟了50个量子位。
The simulation also accounted for the noise that affects real quantum hardware. That builds on work by researchers at Jülich, who simulated 50 qubits on JUPITER in November 2025 under ideal, noise-free conditions.
“朱庇特”于2025年9月投入运行，是欧洲首台百亿亿次超级计算机，每秒可执行超过10亿亿次计算。在最新的测试中，团队在8192块GPU上使用了英伟达GH200芯片。
JUPITER, which opened in September 2025, was Europe’s first exascale supercomputer. It can perform more than a billion billion calculations per second. For this latest test, the team used Nvidia GH200 chips across 8,192 GPUs.
英伟达表示，该模型最初是在其自家量子实验室的一个较小系统上开发和测试的，将工作时间从几周缩短到了几天。
Nvidia first developed and tested the model on a smaller system at its own quantum lab, cutting the work from weeks to days, the company said.
其初衷是帮助研究人员在制造实际硬件之前，测试量子计算机设计的潜在表现。整个模拟完全在传统计算机上运行，而非量子处理器。
The idea is to help researchers test how a quantum computer design might perform before building the actual hardware. The entire simulation ran on conventional computers, not a quantum processor.
英伟达首席执行官黄仁勋表示：“实用的量子计算将需要极其复杂的容错系统。”
“Useful quantum computing will require fantastically complex, fault-tolerant systems,” said Nvidia CEO Jensen Huang.
于利希超算中心主任克里斯泰尔·米歇尔森表示：“欧洲对‘朱庇特’的投资，让研究人员得以探索此前无法触及的问题。”
“Europe’s investment in JUPITER gives researchers the ability to investigate questions that were previously beyond reach,” said Kristel Michielsen, director of the Jülich Supercomputing Centre.
此次宣布之际，黄仁勋正准备于周三在柏林举行的 Nvidia GTC 大会上发表主题演讲。本周早些时候，德国在首都举行的投资峰会上与 Nvidia 签署了一项 AI 合作伙伴关系。
The announcement comes as Huang prepares to deliver his keynote at Nvidia’s GTC conference in Berlin on Wednesday. Earlier this week, Germany signed an AI partnership with Nvidia at an investment summit in the capital.