AMD Medusa Point “Zen 6” APU 跑分再曝，支持 FP16“AVX-VNNI”指令集

IT Home News on March 18 reports that tech media Wccftech today (March 18) published a blog post, claiming that in the GeekBench benchmark database, traces of AMD Medusa Point “Zen 6” APU were found again. The single-core score in version 6.6.0 is 2300 points, and the multi-core score is 13002 points.

Earlier, IT Home reported that the chip appeared on the Plum-MDS1 evaluation platform, using a new FP10 BGA socket designed for the next-generation SoC in the 28-45W power range.

Meanwhile, the latest leaked benchmark records show that the processor is identified as “Ryzen 9,” featuring a 10-core 20-thread design. It uses a hybrid architecture with 4 performance cores and 6 energy-efficiency cores. In addition, it is equipped with 32MB of third-level cache (L3) and 10MB of second-level cache (L2).

Based on data from the benchmark page, the “Zen 6” chip’s base frequency is rated at 2.40GHz, but in actual benchmark testing, its operating frequency stays around 2.0 to 2.1GHz.

In addition, the benchmark page first confirmed that the “Zen 6” architecture supports the FP16 “AVX-VNNI” instruction set, meaning that AMD will provide full FP16 hardware-level compute capability in next-generation SoCs such as Medusa Point.

IT Home Note: FP16 “AVX-VNNI” translates directly to half-precision floating-point vector neural network instruction set. It is a low-level hardware instruction set specifically designed to accelerate AI computations and machine learning tasks, enabling computers to run AI workloads faster and with lower power consumption.