AMD Prepares for Medusa

End of 2025

AMD has released Zen 5 and is not yet talking about Zen 6, which is said to be codenamed Medusa.

According to some new leaks, Medusa is set to release in late 2025, and AMD’s rapid release cadence is said to be designed to keep AMD at the top of the AI ​​spec charts.

According to sources who spoke with the YouTuber

(MLID), several AMD employees have heard of an architecture called Medusa, but opinions vary as to what it will be.

According to one source, this would be a product family that includes desktops, mobile devices, and a Halo mobile product. AMD’s desktop products are codenamed “ridge,” with their mobile components labeled “point” and “halo” for extreme mobile. For Zen 5, we have Granite Ridge and Strix Point. The Strix Halo is not yet released.

AMD’s roadmap shows that it plans to use both TSMC 4nm and 3nm chips in its lineup, but it has only revealed the 4nm chips so far.

TSMC’s 3nm process appears to be the backbone of the Strix Halo and Zen 6 Medusa. The recently announced Zen 5 chips use TSMC’s N4 process, which is a surprise since it was expected to move to 3nm for that architecture after using 5nm for Zen 4.

According to MLID, the move to 3nm was the initial plan, as Strix Point and Strix Halo were initially designed with a 3nm chip architecture. Issues with the chips and delays from TSMC caused that plan to be abandoned, so Strix Point is now a 4nm monolithic chip.

However, Strix Halo could arrive later this year with a 3nm chiplet design to take on Intel’s Lunar Lake, at least in the neural processing unit (NPU) department.

AMD has already stated that Zen 5 will use both 4nm and 3nm chips, so there’s still more to come on the 3nm front from AMD for this generation.

According to MLID, AMD is determined to keep up with Intel when it comes to NPU performance, which has been the driving force behind its semi-strange launch schedule. It launched the Zen 4 8000 series chips for desktop and mobile devices with only an upgraded NPU, likely to stay ahead of Intel’s Meteor Lake chips in the burgeoning AI arms race.

The Meteor Lake NPU is capable of delivering 11 TOPS, while the Zen 4 8000 series XDNA NPU delivers 16 TOPS. This number will increase to 48 TOPS for Lunar Lake and 50 TOPS for Zen 5 Strix Point.

This leak claims that AMD wants to move to a 3nm I/O die for Medusa slash Zen 6, which is a big change since Zen 4 and Zen 5 use TSMC’s 6nm for this chip.

The leaker also claims that Zen 5 Strix Point was supposed to arrive in the first quarter of this year, but was delayed by two years. Strix Point Halo was delayed by almost a year, but it should be able to crush Intel into dust when it does come out.

We still don’t know what AMD has planned for its Halo mobile product or the rest of the Zen 5 Strix Point lineup. The company has only announced two processors so far, so it has a lot more chips to reveal. Intel’s Lunar Lake isn’t available yet. Although it was unveiled in May, it won’t launch until September.