AMD’s RDNA 3.5 graphics architecture will continue to span the APU lineup till 2029, but premium SoCs will feature new RDNA 5 iGPUs.
AMD RDNA 3.5 iGPUs For Mainstream APUs, RDNA 5 iGPUs For Premium APUs, Claims Insider
AMD’s APUs haven’t seen a major architectural innovation since the launch of RDNA 3. While RDNA 3.5 brought a decent upgrade, it was mostly an optimized version of RDNA 3, and AMD has since released its newer RDNA 4 architecture for discrete GPUs. This new architecture offers better AI and RT cores, & also supports FSR Redstone. So there are quite a few nice updates that AMD could go for if it had released RDNA 4 for its APU family.
But AMD’s most recent APU launch, the Ryzen AI 400 “Gorgon Point”, still relies on RDNA 3.5 with higher clocks. Meanwhile, AMD’s Strix Halo and Gorgon Halo “Ryzen AI MAX” families also use the same RDNA 3.5 architecture, but in bigger configurations, offering up to 40 compute units versus the maximum of 16 CUs on the mainstream family.
According to Kepler_L2, it looks like the RDNA 3.5 graphics architecture will be part of AMD’s current and upcoming APU family for a good while. It is stated that AMD APUs will continue to leverage RDNA 3.5 until 2029, which aligns with recent information shared by industry insiders such as Golden Pig Upgrade.
The difference is that after some time, RDNA 3.5 iGPUs will be segmented into the entry-level & mainstream APU designs. AMD’s next-gen Medusa Point “Ryzen AI 500” family is expected to be the last major family to receive the RDNA 3.5 graphics architecture for iGPUs. After this, AMD will skip RDNA 4 entirely & move to RDNA 5.
As per Kepler, the RDNA 5 iGPUs will be featured in the “Premium” category. So expect something like Medusa’s Premium or Halo offerings to feature the new architecture with heightened performance and feature sets.
AMD Ryzen APU “iGPU” Configurations:
| APU Family | Codename | GPU Codename | GPU Compute Units (Max) | TFLOPs |
|---|---|---|---|---|
| Ryzen 2000 | Raven Ridge | Vega | 11 Compute Units | 1.76 TFLOPs |
| Ryzen 3000 | Picasso | Vega | 11 Compute Units | 1.97 TFLOPs |
| Ryzen 4000 | Renoir | Vega+ | 8 Compute Units | 2.15 TFLOPs |
| Ryzen 5000 | Cezanne | Vega+ | 8 Compute Units | 2.04 TFLOPs |
| Ryzen 6000 | Rembrandt | RDNA 2 | 12 Compute Units | 3.40 TFLOPs |
| Ryzen 7000 | Phoenix Point | RDNA 3 | 12 Compute Units | 8.30 TFLOPs |
| Ryzen AI 300 | Strix Point | RDNA 3.5 | 16 Compute Units | 5.93 TFLOPs |
| Ryzen AI MAX 300 | Strix Halo | RDNA 3.5 | 40 Compute Units | 14.85 TFLOPs |
| Ryzen AI 400 | Gorgon Point | RDNA 3.5 | 16 Compute Units | ~8.50 TFLOPs |
| Ryzen AI MAX 400 | Gorgon Halo | RDNA 3.5 | 40 Compute Units | ~15.00 TFLOPs |
| Ryzen AI 500 | Medusa Point | RDNA 3.5+ | TBD | TBD |
| Ryzen AI 500 | Medusa Premium | RDNA 5 | TBD | TBD |
| Ryzen AI MAX 500 | Medusa Halo | RDNA 5 | TBD | TBD |
| Ryzen AI 600 | TBD Point | RDNA 5 | TBD | TBD |
| Ryzen AI MAX 600 | TBD Halo | RDNA 5 | TBD | TBD |
For Medusa, AMD is expected to increase the core counts and add some optimizations in a specific class of APUs to make the iGPU part competitive against Intel’s lineup, which just got upgraded to Xe3, promising a big bump in performance, and far exceeding the capabilities of AMD’s mainstream APU family. Also this year, Intel is expected to roll out Xe3P, a brand new architecture leveraging the Celestial design, under a brand new Arc family.
But Medusa should also be the first family to start equipping brand new RDNA 5 GPU cores. In one post by Kepler at Anandtech Forums, he states that Medusa Premium will feature an RDNA 5 “AT4 GMD” die & Medusa Halo will feature a RDNA 5 “AT3 GMD” die. These dies will be separate from the main compute tiles, and the monolithic version will be Medusa Point, which will retain RDNA 3.5 iGPUs.
In the future, Intel has plans for Xe4 and beyond, while their recent partnership with NVIDIA is also likely to bring a new class of x86 SoCs to the market with RTX-tier graphics options. Intel has made it clear that this partnership doesn’t affect their own graphics product roadmap, so it will be interesting to see how the iGPU landscape changes in the coming years.
Follow Wccftech on Google to get more of our news coverage in your feeds.
