Four Lenovo systems with Medusa Point samples ran Geekbench 7 in early November, covering three core configurations: 10 (4+6), 8 (2+6), and 6 (1+5). The 8-core variant reports 8 MB and the 6-core variant 4 MB.

The GPU identifier GFX1170 confirms an RDNA 4m iGPU, distinct from the full RDNA 4 found in GFX12xx parts. AMD is tiering cache aggressively across the lineup.

The drop from 32 to 16 MB between April and November is a stepping change in engineering silicon, not a spec revision. The older reading came from a different suffix and a different benchmark generation, and the ComputerBase wire page flags the max-clock values as implausible.

AMD is capping the 10-core Medusa Point at two-thirds the cache its 12-core Gorgon Point flagships carry, so the 10-core part will not out-cache the Ryzen AI 9 HX 475.

No boost clock, NPU TOPS figure, power envelope, product name, launch date, or price appears in any of the four entries.

The draft buries the actual news—the cache halving—under a list of four systems. The rewrite leads with the 32-to-16 MB drop, which is the only fact that changes a reader's mental model of the part. Rhythm shifts from a single dense opening to shorter, punchier paragraphs that separate the spec table, the GPU note, the stepping explanation, and the Gorgon Point comparison. The title now names the change rather than describing a Geekbench posting.