A rack of two Hygon Server-based processors was spotted running Geekbench’s new artificial intelligence-based benchmark on Geekbench Browser. Benchmark leak for X The report states that the CPUID name of the chip is 900F22.
Unfortunately for the company, the dual-CPU configuration achieved poor results in Geekbench’s AI benchmark test. The single-precision score is 1,412 points, the half-precision score is 531 points, and the quantization score is 1,523 points. Both CPUs have a base frequency of 3GHz and are paired with 64GB of memory.
By comparison, Intel’s nearly 10-year-old Skylake-based Core i7-6700HQ quad-core mobile CPU is barely slower than the dual 32-core chip, with a single accuracy score of 1,113 points, a semi-accuracy score of 589 points, and a quantitative score of 1,394 points.
| Row 0 – Cell 0 | single precision fraction | half precision fraction | Quantitative score |
| Dual Hygion 16 core CPU | 1,412 | Chapter 531 | 1,523 |
| Core i7-6700HQ | 1,113 | Chapter 589 | 1,394 |
| Ryzen 5 7600X | 3,542 people | 1,686 | 1,394 |
Compared to more modern CPUs, Hygon-based server racks are (expected to) perform far better than mid-range desktop CPU hardware. For example, an AMD Ryzen 5 7600X user report we extracted showed a single-precision score of 3,542 points, a half-precision score of 1,686 points, and a quantized score of 6,281 points, which represents a 2.5 to 4 times performance improvement compared to Hygon server CPUs , albeit at a fraction of the core count.
We can’t confirm exactly what Hygon CPU model was tested, but regardless, Hygon has been using only AMD’s original Zen architecture in all of its server chips since 2017 – in fact, it has found a way to use AMD’s Zen Methods for porting the architecture to AMD’s architecture. SP5 socketpowering its latest Zen 5-based chips. That’s why even though Hygon’s CPU has many cores, its performance is still severely lacking compared to all modern CPUs.
Hygon is a Chinese fabless chip maker that, due to U.S. sanctions, can’t use anything other than AMD’s original Zen architecture (at least for now), which is the company’s Achilles’ heel in the race to improve performance. The only way it can improve performance is by adding more cores to its CPUs, and adding more CPUs to its platform, which does help with multi-core performance, but crucially for most other areas (e.g. single-core IPC) Not helpful and the bottlenecks associated with latency, not to mention missing out on the latest CPU instruction sets (e.g. AVX-512).