AMD Medusa APU Benchmarks Leak: Zen 6 Chips Post Strong Geekbench Scores
AMD’s next laptop chips have shown up early. Benchmark entries for the company’s upcoming Medusa APUs have appeared in the Geekbench database, and the early numbers point to a large single-core jump over today’s Ryzen AI laptop chips. The listings come from Lenovo test systems, which suggests the silicon is already being exercised inside machines aimed at a 2027 launch window.
Three configurations have surfaced so far: a 10-core chip, an 8-core chip, and a 6-core chip. All three mix full Zen 6 cores with smaller Zen 6c cores, and all three are engineering samples, so treat the scores as an early preview rather than a final verdict. Still, the pattern across the results is consistent, and it is good news for anyone waiting on AMD’s next laptop generation.
Three Medusa Configurations Spotted in Geekbench
The leaked entries describe Medusa Point, AMD’s next-generation mobile APU family. Wccftech reports that the chips were found running in Lenovo devices that form part of the PC maker’s upcoming 2027 AMD lineup. Guru3D, which examined one of the public Geekbench 7 results, confirms the 10-core chip carries the engineering sample ID 100-000001713-32_Y and ran on a Lenovo system identified as 22ETPF6DTA.
Here is how the three leaked configurations compare in Geekbench 7, based on the reported results:
| Configuration | Core mix | Family ID | Geekbench 7 single-core | Geekbench 7 multi-core |
|---|---|---|---|---|
| Medusa Point 10-core | 4 Zen 6 + 6 Zen 6c | Family 26 Model 129 | 3100 | 18313 |
| Medusa Point 8-core | 2 Zen 6 + 6 Zen 6c | Family 26 Model 224 | 2913 | 15075 |
| Medusa Point 6-core | 1 Zen 6 + 5 Zen 6c | Family 26 Model 224 | 2651 | 10670 |
The 10-core model was tested with 64 GB of DDR5-5200 memory, according to Wccftech. Guru3D’s reading of the same Geekbench entry lists 55.58 GB of DDR5 running at 5,188 MT/s across four channels, plus 16 MB of shared L3 cache and AVX-512 instruction support. The hybrid layout is the headline here: AMD is pairing a small number of full-power Zen 6 cores with a larger block of efficient Zen 6c cores, a split that should let thin laptops stay cool while still boosting hard on single-threaded work.
Single-Core Gains Lead the Story
The most striking figures are the single-core scores. In Geekbench 6, Wccftech reports the 10-core engineering sample scored 3,407 points in single-core testing and 16,826 in multi-core testing. That single-core result puts it about 15 percent ahead of both the Ryzen AI MAX+ 395 and the Ryzen AI 9 HX 470, two of AMD’s current flagship laptop chips, even though both of those run at clock speeds above 5.00 GHz.
Geekbench 7 tells a similar story, and the margins grow wider:
- 10-core Medusa: 3,100 single-core and 18,313 multi-core. Wccftech reports it is 28 percent ahead of the Ryzen AI MAX+ 395 and 32 percent ahead of the Ryzen AI 9 HX 370 in single-core testing.
- 8-core Medusa: 2,913 single-core and 15,075 multi-core. That matches the 12-core Ryzen AI 9 HX 370 in multi-core testing, despite having four fewer cores.
- 6-core Medusa: 2,651 single-core and 10,670 multi-core. Wccftech reports it is 25.1 percent faster than the 6-core Ryzen AI 5 340 in single-core testing and up to 18 percent faster in multi-core testing.
The multi-core picture is strong too. The 10-core Medusa sample beats the 12-core Ryzen AI 9 HX 370 by 21.5 percent in Geekbench 7 multi-core testing, according to Wccftech, and comes close to the 16-core Ryzen AI MAX+ 395. For a 10-core engineering sample to trade blows with a 16-core flagship is the kind of result that gets laptop makers’ attention, and it hints that Zen 6 brings a real per-core improvement rather than a small clock bump.
For context on how current AMD flagship desktop chips compare, our Ryzen 9 9900X3D vs 9950X3D gaming comparison breaks down where the extra cores actually matter and where they do not.
The RDNA 4m Graphics Core: GFX1170
The leak is not only about the CPU cores. The same Geekbench database entries include an OpenCL result for a graphics device identified as GFX1170, which Wccftech reports is the RDNA 4m integrated GPU inside the 10-core Medusa chip. The “m” in RDNA 4m points to a mobile-focused design that fuses RDNA 3.5 and RDNA 4 architectural blocks, and this particular sample carries just four compute units.
Even with only four compute units, the reported OpenCL score of 22,249 points is notable. Here is how it stacks up against current Radeon laptop graphics, based on the comparison figures reported by Wccftech:
| Integrated GPU | Compute units | OpenCL score |
|---|---|---|
| Radeon 890M | 16 CU (RDNA 3.5) | 31475 |
| Radeon 880M | 12 CU (RDNA 3.5) | 26814 |
| Radeon GFX1170 (RDNA 4m) | 4 CU | 22249 |
| Radeon 860M | 8 CU (RDNA 3.5) | 20642 |
| Radeon 840M | 4 CU (RDNA 3.5) | 12019 |
Read that table carefully. The 4-CU RDNA 4m part scores nearly double the 4-CU Radeon 840M, and it edges past the 8-CU Radeon 860M. It still trails the 12-CU 880M and the 16-CU 890M, but those are much larger graphics blocks. The interesting question is what happens when AMD ships Medusa chips with more RDNA 4m compute units, because the per-unit performance looks like a clear step up from RDNA 3.5.
That matters for laptop gaming. If the integrated graphics in a mainstream 2027 laptop can handle light and medium gaming on its own, fewer buyers will need a discrete GPU at all. Readers who are building a budget gaming PC today should still buy a discrete card, since integrated graphics cannot match a dedicated GPU yet, but the gap keeps shrinking with each APU generation.
Read These Numbers With Care
Before anyone treats these scores as a buying guide, the caveats matter as much as the numbers. Guru3D flags the biggest one: the Geekbench entry lists a nominal 2.00 GHz base clock and a maximum frequency reading of just 1,996 MHz, which is unusually low for a chip posting these scores. That reading is a warning against treating this benchmark run as representative of final clock speeds, and Guru3D notes that power limits, firmware, cooling, memory configuration, and incomplete silicon can all move a pre-release result.
A few more cautions to keep in mind:
- These are engineering samples, not retail chips. Performance can change in either direction before launch, and Wccftech itself notes that retail samples should be faster, though that is an expectation, not a guarantee.
- AMD has confirmed nothing. Guru3D states plainly that AMD has not confirmed the product name, configuration, clocks, cache, performance, or release timing. Medusa is a codename from leaked roadmaps and benchmark entries, not an official announcement.
- One benchmark is not a review. Geekbench measures a narrow slice of CPU work. Real gaming performance depends on thermals, power limits, drivers, and the graphics block, none of which a single database entry can settle.
- The laptop is unknown. The Lenovo system identifier proves a development platform exists, but it does not identify a final commercial model or the power envelope AMD and Lenovo are targeting.
None of that makes the leak meaningless. A chip does not appear in a public benchmark database with a 10-core hybrid layout unless the silicon is real and running in systems. The scores are best read as a direction of travel: Zen 6 looks like a meaningful step up in per-core performance, and RDNA 4m looks like a meaningful step up per compute unit.
What Medusa Could Mean for Laptop Gamers
If the leaked scores hold up in retail silicon, Medusa Point shapes up as a strong mainstream laptop chip for 2027. The hybrid core layout fits how most people actually use a laptop: a few fast cores for the game or app in focus, and efficient cores handling everything else in the background. The single-core gains should show up in everyday responsiveness, game loading, and the many PC games that still lean on one or two threads.
The graphics side may matter even more for casual players. A 4-CU RDNA 4m part already beating the 8-CU Radeon 860M suggests that a future Medusa chip with 8 or 12 compute units could run esports titles and older AAA games at 1080p without a discrete GPU. That would make thin and light laptops far more capable gaming machines, and it would put pressure on entry-level discrete graphics.
There is also a software angle worth watching. AMD keeps adding frame generation and upscaling support across its hardware, and our guide to AMD’s frame generation tech on RDNA 3 and RDNA 4 cards shows how much extra smoothness those features can add on supported hardware. If RDNA 4m carries the same feature set into laptops, Medusa machines could punch above their weight in frame rates, not just in raw benchmark scores.
What to Watch Next
The next milestones are straightforward. First, more Geekbench entries should appear as Lenovo and other laptop makers test additional Medusa configurations, including any higher-end SKUs with more graphics compute units. Second, retail samples will eventually replace engineering samples in the database, and those scores will tell us far more about the final product. Third, AMD will need to put a name and a launch window on the record, most likely closer to the 2027 laptop refresh cycle.
Until then, the Medusa leak is an encouraging early signal, not a promise. The Zen 6 cores look fast, the RDNA 4m graphics look efficient, and the silicon is already running in real Lenovo test machines. For PC gamers shopping for a laptop in 2027, that is a genuinely good sign. For everyone else, the sensible move is to wait for retail silicon and independent testing before drawing conclusions.
News Source: Wccftech, with corroborating detail from Guru3D. All benchmark figures in this article are leaked engineering-sample results reported by those sources, not our own testing.
