Loongson 3B6000: Китайский «кремниевый суверенитет» разбился об AMD

Politics often breaks against physics, and Chinese microelectronics proved this once again. While Beijing reports successes in import substitution, the dry numbers of benchmarks paint a far less rosy picture. The latest Loongson 3B6000 processor, which was supposed to be an answer to Western giants, lost to the mass-market AMD Ryzen chip by almost three times in real tests. This is not just a slight lag—it's a chasm that cannot be bridged by a couple of patches or patriotic slogans.

Loongson's history did not begin yesterday. China spent years trying to build its own ecosystem, initially relying on the MIPS architecture, then creating its own—LoongArch. The idea was ambitious: to completely eliminate dependence on x86 and ARM by creating a "third way." One might have thought that the 3B6000 would become a triumph of this strategy. Chinese engineers claimed significant increases in IPC, the performance per clock cycle, promising a level not quite of tomorrow but at least of a solid present. However, reality proved cruel to the ambitions of the Middle Kingdom.

When it came to direct comparison with the Ryzen 5 9600X, it turned out that the Chinese chip cannot even handle basic loads at the level of its competitor. A threefold difference means that to perform the same task, a Chinese server or workstation would need either three times more time or three times more processors. In a world where computational efficiency directly converts to money and the speed of neural network training, such an arrangement looks catastrophic. The problem here is not only in the architecture but also in access to modern processes, which for China is now maximally difficult.

Why is this important right now? We are in the midst of an arms race in the field of artificial intelligence. Working with large language models requires not only powerful GPUs but also fast central processors capable of efficiently managing data streams. If the "heart" of the system, the CPU, works slowly, then even the fastest accelerators will sit idle. Chinese companies such as Baidu or Alibaba are forced to build their cloud systems based on this hardware, which automatically puts them at a disadvantage compared to Google or Microsoft.

Moreover, the architectural isolation of LoongArch creates a "chicken and egg" problem. Software developers are in no hurry to optimize their products for the Chinese processor because it has a small market share, and users do not buy the processors because everything runs slowly on them. Without compatibility with the huge body of software accumulated for x86, Loongson remains a niche product for government structures, where the origin of the chip matters more than its actual power. This creates a situation where a country lives in a technological bubble that is increasingly distancing itself from the world frontier.

China's attempt to leap over decades of development by the Western semiconductor industry in just a few years looks heroic, but so far has borne no fruit. It turns out that simply drawing a chip schematic is not enough—one needs access to lithography, unique materials, and most importantly, access to the global market of talent and ideas. Loongson 3B6000 became a vivid lesson in how sanctions and isolation slow progress, turning once-promising developments into lagging shadows of the past.

The key question: Will China be able to compensate for its technological backwardness through scaling, or will its AI industry forever remain hostage to slow "sovereign" silicon?