AMD has clearly dropped the ball when it comes to competitiveness in the mobile notebook marketplace. Gorgon Point was a slight speed update of Strix Point, and increasing the maximum boost frequency by 100 to 150 MHz for the CPU and the Radeon 890M from 2.9 GHz to 3.1 GHz didn’t really move mountains.
The Strix Halo sits on a throne of its own and serves an interesting, AI-heavy part of the market.
AMD will announce Zen 6 for the data center, codenamed Venice, at the Advancing AI event on July 22, but the client solutions are expected to launch at a later date.
It looks like AMD’s upcoming Zen 6 mobile architecture, codenamed Medusa Point, is shaping up to pack a serious punch, even in its early engineering stages.
A fresh entry in the Geekbench database shows a 10-core Medusa Point APU comfortably outperforming the current flagship Ryzen AI 9 HX 370 by a notable margin. However, the Ryzen AI 9 HX 470 from Lenovo—simply named LENOVO 21VF—is scoring 2,986 in single-thread and 16,407 in multi-thread performance. This is a 12-core system, while the leaked Medusa Point processor runs 10 cores.
The leaked scores for Medusa Point bring single-thread performance to 3,174 and multi-core to 15,092.
The benchmark popped up under the familiar “AMD Plum-MDS1” platform evaluation board, featuring an engineering sample chip designated as 100-000001713-33_N. The software detects it as a 10-core, 20-thread part, armed with 10MB of L2 cache and 32MB of L3 cache.
Significant jump over early silicon
Back in March, an earlier revision of this exact same silicon hit the database with rather disappointing numbers. This time around, the updated testing shows a massive performance optimization.
The chip put up a single-core score of 3,174 points and a multi-core score of 15,092 points in Geekbench 6.6.0 (running the Windows AVX2 test). Compared to the March leak, that is a massive 38 percent leap in single-thread performance and a 16 percent increase in multi-core metrics.
Leaving Strix Point behind
What makes this interesting is how it stacks up against the current top-tier Ryzen AI 9 HX 370 (Strix Point). The current flagship at its highest iteration, with the Asus ProArt PX13, scores an impressive 2,951 in single-core and 15,867 in multi-core.
This means the early Zen 6 sample is already beating its predecessor by a modest 7.5 percent in single-threaded tasks and is 5 percent slower in multi-threaded workflows, mostly due to the fact that it has fewer cores (10) than the 12-core HX 370.
Geekbench currently clocks the chip at a 2.0 GHz base and a 2.06 GHz maximum frequency, which we can safely ignore as a reporting error common with early silicon, as a 2 GHz x86 clock wouldn’t yield a 3,100+ single-core score unless AMD invented some magical IPC space tech.
The leak also confirms expected instruction sets, including AVX512-FP16 and AVX-VNNI support. On the graphics side, rumors point toward this chip carrying an updated “RDNA 4m” iGPU, which combines RDNA 3.5 graphics blocks with the enhanced AI ray-tracing features of RDNA 4.
At the final stage, I would imagine that Medusa Point will also get a 12-core version, if not even more, which will be more competitive in multi-core results and expectations.
We don’t expect to see Medusa Point on store shelves until late next year, but the preliminary architecture performance looks quite promising.
