AMD Medusa Halo APUs leak reveals LPDDR6 support with 80% bandwidth boost for Ryzen AI MAX 500

AMD's future 'Medusa Halo' APUs could use LPDDR6 RAM -- new leak suggests Ryzen AI MAX 500 series could have 80% more memory bandwidth

Following its well-received family of Strix Halo APUs, AMD will refresh its high-end gaming APU lineup with Gorgon Halo, bringing it up to date with the recently announced Gorgon Point, aka Ryzen AI Max 400, series. The real fun may arrive after that, though, as the true next-gen upgrade will come in the form of "Medusa Halo" sometime in 2027-28. This purported high-end APU could feature Zen 6 CPU cores and RDNA 5 graphics. Now, a new leak from reputed leaker Olrak29_ says that it will also support LPDDR6 memory. The incredibly descriptive post above might only have one word in it, but the inclusion of the LPDDR6 standard on a potential "Ryzen AI Max 500" series promises some great upgrades. Right now, Strix Halo features a 256-bit wide LPDDR5X memory controller that supports 8,000 MT/s speeds, which results in a bandwidth of 256 GB/s. The Gorgon Halo refresh could increase that figure to 273.1 GB/s with its 8,533 MT/s config. When we bring Medusa Halo into the picture with LPDDR6 -- even with the same 256-bit memory bus -- at 14,400 MT/s, we're looking at 460.8 GB/s of maximum throughput, around 80% higher than current-gen Strix Halo. Prior rumors have alleged that the bus width will be upgraded to 384-bit on Medusa Halo, which would result in an insane 691.2 GB/s of memory bandwidth. Higher memory bandwidth on powerful APUs like the Halo series is of ever greater importance as AMD targets its Halo series not only at gaming but also for AI applications. LLM inference relies on memory bandwidth for high performance, which makes the raw throughput of the shared memory a very important characteristic to look out for. And Medusa Halo seems like it won't disappoint here. AMD's competitors aren't sitting still in this regard. Intel's Panther Lake features the fastest x86 memory controller at the moment with its support for LPDDR5X-9600, although Intel has publicly disavowed building an integrated graphics processor as large as Strix Halo's. Meanwhile, Apple's M-series SoCs top out at 819 GB/s on the M3 Ultra with its 1,024-bit interface. Take all of these numbers with a grain of salt, however because AMD hasn't even confirmed a Gorgon Halo refresh, let alone Medusa Halo; that product family is likely a couple of years out at this point. The company just unveiled two new Strix Halo SKUs at CES last month, so even Strix Halo is not done yet. Roadmap leaks so far have said AMD will maintain RDNA 3.5 graphics across most of its lineup through 2027-28, but Medusa Halo could be an exception.

AMD Medusa Halo "Ryzen AI MAX" SoCs To Feature LPDDR6 Memory Support

AMD Medusa Halo "Ryzen AI MAX" SoCs To Elevate Bandwidth Performance With LPDDR6 Memory AMD's Ryzen AI MAX series saw two new additions at this year's CES, and the company is already working on the refresh for its first-gen Halo lineup. The first-generation Strix Halo series already maintains a super-premium position as an SoC for AI PCs, and with Gorgon Halo (Ryzen AI MAX 400), AMD will continue to push performance further with boosted CPU/GPU clocks and higher memory speeds. As for the true successor to Strix Halo, that will come in the form of Medusa Halo or Ryzen AI MAX 500 series. These will be scheduled sometime in 2027-2028, and will feature brand new Zen 6 CPU cores and RDNA 5 GPU cores. As per rumors, AMD will have a diverse set of Medusa SoCs, with entry-level SoCs getting RDNA 3.5 iGPU cores while more premium and Halo SoCs will get newer RDNA 5 iGPU configs. Although it's too early to tell what kind of specs we will get with Medusa Halo, up to 24 cores, and either similar or beefier GPU core configs, it looks like the platform itself will take advantage of the next-gen LPDDR6 memory standard. We know from JEDEC's own confirmation that LPDDR6 memory will feature speeds of 14,400 MT/s across a 24-bit wide channel, and with 38.4 GB/s bandwidth per module. Strix Halo initially featured up to 8000 MT/s support, with up to 256 GB/s bandwidth. This will be updated to 8533 MT/s in the upcoming refresh lineup. But with Medusa Halo, even with the same bus-width of 256-bit, we are looking at a bandwidth of 460 GB/s, a 80% uplift. This would prove to be a major uplift for the onboard GPU, resulting in far better performance. Currently, Intel's Panther Lake 12Xe3 SoCs feature the fastest LPDDR configuration of up to 9600 MT/s on notebook PCs. While Intel is building up its own SoC roadmap with powerful offerings and working with NVIDIA, AMD remains unfazed by such partnerships and has already stated that it will continue to offer disruptive technologies in the future. This would make AMD Ryzen AI MAX 500 "Medusa Halo" a powerful SoC, building on the foundation of Strix and Gorgon Halo, which are disruptive offerings for the AI PC and gaming segment. AMD's recently introduced Ryzen AI Halo platform will also receive updates with each respective generation of Halo SoCs.