A recent Geekbench AI benchmark has revealed surprisingly poor performance from a Chinese-made Hygon server system featuring two 16-core processors, barely matching the capabilities of Intel's aging mobile chip from nearly a decade ago.
The dual-CPU Hygon server configuration, identified with CPUID 900F22, achieved modest scores of 1,412 points in single precision, 531 in half precision, and 1,523 in quantized tests. The system ran at 3GHz with 64GB of memory.
In a striking comparison, Intel's quad-core processor from 2015 delivered nearly identical performance, scoring 1,113 points in single precision, 589 in half precision, and 1,394 in quantized tests.
The performance gap becomes even more apparent when comparing to modern processors. AMD's Ryzen processor, a mid-range desktop CPU, outperformed the Hygon server system by 2.5 to 4 times across all tests, despite having fewer cores.
The underwhelming results stem from Hygon's reliance on AMD's original Zen architecture from 2017. Due to U.S. trade restrictions, the Chinese chipmaker cannot access newer AMD architectures, limiting its ability to improve single-core performance and implement modern instruction sets like AVX-512.
While Hygon has managed to adapt the old Zen architecture to work with AMD's latest SP5 socket, the company's only path to increased performance is adding more cores or processors to their systems. This approach helps with multi-core performance but fails to address fundamental limitations in single-core efficiency and latency.
This benchmark highlights the growing technological gap between Chinese domestic processors and contemporary Western chips, as trade restrictions continue to impact China's semiconductor industry.