A rack comprised of two Hygon server-based processors was seen on the Geekbench browser that features a run of Geekbench’s new AI-based benchmark. Benchleaks on X experiences that the chip’s CPUID title is 900F22.
Sadly for the company, the dual-CPU configuration scored abysmal ends in Geekbench’s AI benchmark. The one precision score was 1,412 components, half precision score 531, and the quantized score was 1,523. The two CPUs operated at a base frequency of 3GHz flat and have been paired with 64GB of memory.
By comparability, Intel’s virtually 10-year-old Skylake-based Core i7-6700HQ quad-core cell CPU was barely any slower than the dual 32-core chips, that features a single precision score of 1,113 components, half-precision score of 589 components and a quantized score of 1,394 components.
Row 0 – Cell 0 | Single Precision Ranking | Half Precision Ranking | Quantized Ranking |
Twin Hygion 16-core CPUs | 1,412 | 531 | 1,523 |
Core i7-6700HQ | 1,113 | 589 | 1,394 |
Ryzen 5 7600X | 3,542 | 1,686 | 1,394 |
As compared with a far more modern CPU, the Hygon-based server rack is (expectedly) vastly outperformed by mid-range desktop CPU {{hardware}}. As an illustration, one AMD Ryzen 5 7600X particular person report we pulled up scored 3,542 components in single precision, 1,686 components in half-precision, and 6,281 components in quantized, representing a 2.5x to 4x effectivity enchancment over the Hygon server CPUs, no matter having solely a fraction of the number of cores.
We can’t guarantee what precise Hygon CPU model was examined, nevertheless regardless, Hygon solely makes use of AMD’s distinctive Zen construction from 2017 in all of its server chips — in-fact, it has found a way to port AMD’s Zen construction into AMD’s latest SP5 socket, which powers its latest Zen 5-based chips. That is the explanation Hygon’s CPUs endure a extreme effectivity deficit compared with all modern CPUs, no matter having many cores.
Hygon is a fabless Chinese language language-based chip maker, and on account of U.S. sanctions, it could presumably’t use one thing previous AMD’s distinctive Zen construction (a minimal of for now), which is the company’s Achilles heel for aggressive generational effectivity enhancements. The one means it has been able to extend effectivity is by together with additional cores to its CPUs and together with additional CPUs to its platform, which does help improve multi-core effectivity nevertheless critically doesn’t help in most totally different areas corresponding to single-core IPC and latency-related bottlenecks, to not level out missing out on the latest CPU instruction models corresponding to AVX-512.