AMD Zen 6 benchmark leak: 2GHz vs 5GHz, single-core ahead by 29%, multi-core ahead by 22%

AMD Zen 6 Benchmark Leak: 2GHz vs 5GHz, Single-Core Beats by 29%, Multi-Core by 22% — The Geekbench database has been buzzing again lately, as an unreleased chip has managed to outpace the current flagship. AMD's next-generation Medusa Point — an engineering sample of the APU based on the Zen 6 architecture — has run another round of benchmarks. This time, the data is packed with insights.

💡 What You Will Learn

AMD Zen 6 Benchmark Leak: 2GHz vs 5GHz, Single-Core Beats by 29%, Multi-Core by 22% — The Geekbench database has been buzzing again lately, as an unreleased chip has managed to outpace the current fla

AMD Zen 6 Benchmark Leak: 2 GHz vs 5 GHz, Single-Core Up 29%, Multi-Core Up 22%

The Geekbench database has been buzzing again, with an unreleased chip putting a gap between itself and the current flagship.

AMD's next-gen Medusa Point—an APU engineering sample based on the Zen 6 architecture—has run another round of benchmarks. This time, the data is packed with insight, not because it scored high, but because it scored remarkably "low" while still delivering results that nail the current flagship to the wall.

Not the First Time in the Spotlight

Medusa Point actually first appeared on Geekbench back in March. That sample used the Plum-MDS1 evaluation platform, featuring the new FP10 BGA socket—the foundation for future 28-45W laptop APUs. Back then, it ran at a 2.4 GHz base clock, dropping to around 1.3 GHz during actual runs—classic early engineering sample behavior with voltage and frequency not yet tuned. Even so, it already managed to beat Strix Point in multi-core performance.

Four months later, the chip has resurfaced. The clock dropped from 2.4 GHz to 2.0 GHz, but it's running more stable, and the scores have jumped significantly. This suggests AMD's engineering team hasn't been idle—microarchitecture tuning is progressing steadily, and the core design is converging quickly.

Interestingly, CPUID's HWMonitor mentioned "preliminary support for Medusa Point" in its changelog late last year, indicating the chip's engineering progress was ahead of what outsiders expected.

A 2 GHz Chip Obliterating a 5 GHz Chip

Let's look at the numbers.

This Medusa Point is a 10-core, 20-thread configuration, with 4 performance cores and 6 efficiency cores. Cache-wise, it packs 32MB of L3 and 10MB of L2.

This 4+6 hybrid core setup is intriguing. AMD didn't simply copy Intel's P-core plus E-core approach; it's more like two optimized branches on the same Zen 6 architecture: performance cores focused on single-core speed and cache, while efficiency cores handle multi-core throughput and power draw. The efficiency cores aren't stripped-down "little cores"—they're density-optimized versions of the same architecture. That's a fundamental difference in AMD's and Intel's approaches.

What's particularly striking is the clock speed. This sample has a base clock of just 2.0 GHz, peaking at 2063 MHz. Its comparison target, the Strix Point (Ryzen AI 9 365, Zen 5 architecture), can boost up to 5 GHz—more than double the frequency difference.

And the results? Check the table:

2 GHz beating 5 GHz, winning single-core by 30%. That kind of data on the table is frankly a bit absurd.

What This Engineering Sample Really Means

A lot of people's first reaction to "2 GHz" is: it's running at such a low clock, it must still be early days.

That's true. But the real value of this data isn't how strong Medusa Point is right now—it's what it reveals about the Zen 6 architecture's underlying potential.

Geekbench 6 scores, at the same clock speed, essentially measure IPC—instructions per clock cycle. If a Zen 6 sample at 2 GHz can match or even surpass Zen 5 at 5 GHz, that implies its IPC could be 40-50% higher than Zen 5.

Of course, this is an extreme extrapolation; production versions will balance clock speed, power, and die area. But the direction is clear: Zen 6 isn't a minor refresh—it's a major leap.

Zen 6 Isn't Just Faster—It Has New Tricks

This benchmark run also exposed a technical detail: Geekbench reported for the first time that Zen 6 supports FP16 operations in the AVX-VNNI instruction set.

To put it simply: previously, when doing AI inference, FP16 operations either had to be emulated (halving performance) or routed through a separate NPU. Now, Zen 6 provides full FP16 support directly at the CPU level—meaning thin-and-light laptops running local AI models can rely on the CPU itself without waiting for NPU scheduling every time.

This is especially important for an APU like Medusa Point. It already has RDNA 5 integrated graphics and an XDNA AI engine; now the CPU side also has FP16 covered, so all three modules can accelerate AI workloads. AMD has set a target of 10x AI performance improvement by 2027, and Medusa Point is the flagship product carrying that ambition.

2026 to 2027: AMD's Product Cadence

Placing Medusa Point in AMD's broader roadmap makes things clearer.

In 2026, AMD's main focus is Gorgon Point—essentially a refined Zen 5 version of Strix Point, with the same architecture but polished clock speeds and efficiency. That's a transitional product.

The real generational shift comes in 2027.

Medusa Point is the APU for mainstream 28-45W laptops, using the new FP10 socket, with Zen 6 plus RDNA 5. Above it sits Medusa Halo—targeting high-performance gaming laptops and workstations, with more cores and a higher power ceiling. Further up, the desktop-side Olympic Ridge is expected to use TSMC's 2nm process while maintaining AM5 socket compatibility. 2027 will be a major year for AMD with simultaneous generational leaps in architecture, process node, and graphics.

One more note: this sample ran Windows, not some Linux-tuned benchmark. Its reference value is far more realistic than certain lab-specific data.

See You in 2027

According to AMD's current roadmap, Medusa Point is expected to debut in 2027, with more details at CES 2027. Before that, we'll see Gorgon Point roll out to the market in the second half of 2026.

Zen 5 just launched not long ago, and now Zen 6's formidable hand is already revealed. All I can say is, AMD clearly isn't letting go of the architecture iteration card.

When Medusa Point actually hits the market, what do you think Intel will bring to the fight? Feel free to share your thoughts in the comments.