Leaked AMD Medusa Point 'Zen 6' 10-core test sample matches current generation equivalent at less than half the clock speed

A series of new reports from the hardware scouting bot, Benchleaks, on X (formerly Twitter), has outed a 10-core engineering sample of AMD's next-generation Zen 6 mobile processors, codenamed Medusa Point. According to Geekbench data, the test chip matched the performance of its current-generation counterpart, the Ryzen AI 9 365, albeit at less than half the clock speed.

These tests were conducted on the Plum-MDS1 system. This is likely the internal codename for the validation and testing platform for AMD's next-generation Medusa Point laptop processors. While the leak does not explicitly mention the processor's name, as is the case with early test samples, it is identified by the OPN (Ordering Part Number) "100-000001713-21_N". These OPN codes serve as a fingerprint for unreleased silicon for logistics and customs manifests.

In terms of specifications, the unreleased chip is listed with 10 CPU cores, divided into two clusters. While the Geekbench metadata explicitly identifies two core clusters, this is likely a reporting error, based on initial reports from hardware sleuth HXL. Much like with Intel's Meteor Lake and Panther Lake, the software appears unable to distinguish a three-tier topology, seemingly packing the four Zen 6c cores and two 'LP' cores into a single six-core block (Cluster 2). This also suggests we are dealing with a Ryzen 5/7 configuration, making this processor a successor to AMD's Kraken Point family.

2

Additionally, the chip features 10MB of L2 cache and 32MB of L3 cache, a solid 33% uplift over the 24MB of L3 cache offered by the incumbent Ryzen AI 9 365, a similar 10-core counterpart which uses a split 16MB + 8MB layout. With Strix Point, AMD opted for a dual-CCX design, where each core type had access to a separate L3 cache pool. This incurred a latency penalty if a thread had to move across the CCX. We're yet to see whether AMD will opt for a similar design philosophy with Medusa Point. If the L3 cache is unified, the dense (Zen 6c) cores effectively get access to 4x as much cache as the Zen 5c cores did.

Running Geekbench 6.6.0 under Windows 11, the CPU can dish out 2,300 points in the single-core test and 13,002 points in the multicore test. To put those numbers into perspective, they effectively match the current-generation Ryzen AI 9 365 (Strix Point), which typically averages around 2,480 points and 12,445 points, respectively. The true bombshell, however, lies in efficiency: Geekbench reports a frequency of just 2 GHz for the Medusa sample, a pale shadow of Ryzen AI 9 365, which can boost to 5 GHz.

That being said, we must approach these CPUs with reasonable expectations. While a 2 GHz chip matching a 5 GHz one makes for stunning headlines, we have to consider that software can misreport frequencies on unreleased hardware. Furthermore, engineering samples like these are never meant to break frequency barriers; their sole purpose is to validate the stability of freshly taped-out silicon.

Following AMD's typical two-year processor cadence, Zen 6 should debut sometime in late 2026, though leaks suggest the consumer-facing Medusa Point and Olympic Ridge families are pushed to 2027, likely coinciding with Intel's Nova Lake family of processors.