At AMD’s Financial Analyst Day conference, the company presented its long-term financial model, strategy, and related matters. More importantly, the roadmaps, i.e., plans for future products, were unveiled as part of the event. We already wrote about the CPU roadmap, which for the first time featured the Zen 7 core. Alongside it, the company also showed a plan for new Radeon GPU architectures and Ryzen mobile processors.
Next-gen Radeon GPUs on the way, remain nameless for now
Unfortunately, we don’t have much concrete information on the GPU side yet. The company’s roadmap currently shows only one next generation following the Radeon RX 9000 series and their RDNA 4 architecture. However, AMD is not revealing what this successor generation will be called—for the graphics cards themselves, this makes sense, as after the RX “9000” series, some change in naming is needed, whether it’s going to be a transition to five-digit numbers or something else. The new naming scheme will likely be revealed by the company only shortly before launch.
But we also don’t know what the successor architecture will be called—it’s not stated whether it will be RDNA 5 or if there’s going to be a rebranding to UDNA, as reported some time ago. According to the Kepler_L2 leaker, AMD has perhaps not yet decided what to call the architecture. It might not be either of the mentioned options, as UDNA is reportedly somewhat of a working name and may not be the actual final designation. For our purposes, we will probably continue to refer to it as RDNA 5 for now, but remember that this name could ultimately be changed and is by no means confirmed.
New ray tracing and AI acceleration
The so-called RDNA 5 architecture is supposed to bring progress corresponding to a generational transition in both AI capabilities and ray tracing acceleration performance. Both received significant upgrades already in RDNA 4, even though rumors circulated for a while that this architecture would be more of a mere “RDNA 3 bugfix”. In the end, it turned out to be a significant leap. According to this roadmap, it seems RDNA 5 might also not be any sort of a mild evolution, but the architecture could again feature a new generation of Ray Accelerators and hopefully significantly better performance in ray-traced games.
The news about a “next-generation AI” in RDNA 5 is also positive, but it’s not specified what exactly this means. Theoretically, AMD could move AI acceleration from the general-purpose CU units to specialized matrix cores found in the CDNA architecture-based Instinct data center GPUs. This would make sense if this is indeed the GPU architecture enabling the convergence of AMD’s data center GPU line for AI (Instinct / CDNA architectures) and gaming GPUs (Radeon, RDNA architecture line). But this is not confirmed anywhere in the roadmap, so the matter remains the subject of speculation. It’s entirely possible that AI will still be accelerated within the CUs.
Some time ago, alleged (and still approximate and non-final) parameters of the planned RDNA 5 GPUs surfaced, which you can find in the following article. However, anything there is not official information yet; AMD has not provided any specifications.
- Read more: RDNA 5 GPU specs leaked. RX 10000 Radeons back in high-end?
- Read more: Details on Cheaper RDNA 5 GPUs: Will Radeons Switch to LPDDR6?
Data center GPUs: Instinct MI500
AMD places a bit more emphasis on the Instinct GPUs for the data center segment, where it promises annual new generation releases. Unlike with Radeons, the company confirmed that the Instinct MI400 accelerators, launching next year (2026) and which we have written about before, will use an architecture named CDNA 5. The MI400 is supposed to have higher raw performance, which will also be contributed to by support for reduced-precision computations (with the FP4 data type). The MI400 is also said to bring HBM4 memory with higher capacity and bandwidth compared to the previous generation.
AMD revealed that it will be followed in 2027 by the Instinct MI500 with the “CDNA Next” architecture, which likely means CDNA 6. So, if AMD continues to stick with the CDNA architecture naming, that could mean that the RDNA naming for gaming GPUs could also remain. By this stage, the CDNA 6 architecture should be transitioning to a foundation based on technologies from the RDNA architecture line.
The Instinct MI500 is supposed to introduce a new generation of compute units, memory, and interconnect logic, but no further details were given.
Ryzen Medusa Point and Medusa Halo APUs with Zen 6 architecture confirmed
In the area of Ryzen processors, AMD confirmed the next generation of “APU” processors, which is means primarily mobile SoCs for laptops, though these processors could eventually also come to the desktop AM5 platform. If you follow us regularly, you have probably already noticed that information about mobile SoCs (APUs) with the Zen 6 architecture, which were allegedly codenamed Medusa Point and Medusa Halo, had previously leaked through unofficial channels. AMD has now officially acknowledged them for the first time in the roadmap, which you can see below.
The confirmation of their Medusa codename cold be a sign that the other leaked information and alleged parameters might be true. However, AMD has not officially communicated anything else yet. Beyond the use of Zen 6 (and Zen 6c) cores which is something reasonable to assume, we don’t know details like what the integrated GPU architecture will be. The manufacturing process node was also not confirmed (while according to unofficial leaks, the chips are supposed to use TSMC’s 3nm technology).
Next year will bring only a Zen 5 refresh
The bad news, however, is that the roadmap in this case confirms the approximate launch timeframe—according to the graphic representation, it’s a product that will not arrive until 2027. So, we will likely not see Zen 6 cores in mobile Ryzens next year, although this might not be entirely final verdict yet.
You might notice that in the graphic, AMD only shows Ryzen 8000 Hawk Point processors for year 2024, and lists the Ryzen AI 300 Strix Point as a 2025 product, even though they managed to launch in the summer of 2024. Theoretically, something similar could happen with the Medusa Point/Medusa Halo APU family, meaning some models might still make it to market in 2026. But AMD isn’t promising that, so the more pessimistic interpretation pointing to 2027 is perhaps more likely.
In any case, AMD has a refresh of the Ryzen AI 300 series marked for 2026, designated as “Gorgon Point”, which is meant to fill the time until the Medusa generation APUs are released. Gorgon Point likely won’t be new chips, but just a revision of the original Strix Point and Krackan Point APUs.
It will therefore retain the same TSMC 4nm node, the same Zen 5 and Zen 5c cores, and the same RDNA 3.5 generation integrated GPU. According to the slides, Gorgon Point will apparently have higher NPU performance (via higher clock speeds), but CPU clock speeds probably won’t increase much. For significant jumps in general performance and efficiency, we’ll have to wait until 2027, and the same will likely apply to the gaming performance of the integrated GPU. Some preliminary parameters for these refreshed APUs have already leaked, which we discussed here:
The Gorgon Point APUs should also finally come to the desktop for the AM5 socket, but we don’t know when that will be yet. The launch of the mobile version of Gorgon Point should probably be sometime early in the year, likely at CES 2026.
English translation and edit by Jozef Dudáš
⠀
