AMD Reportedly Developing Radeon RX 9000 GPU with 32GB VRAM

AMD is working on a new Radeon RX 9000 GPU with 32 GB of VRAM. According to information from the Chiphell forum (thanks VideoCardz), this model could be a version of the RX 9070 XT but with double the memory capacity of the 16 GB version already available for retailers and partners since December.
Possible Design and Architecture
Although AMD has not officially confirmed this GPU, the rumor suggests that it will not be part of the Radeon PRO series, which is aimed at professional users. Instead, it will likely be a gaming GPU and could also be used for advanced applications such as artificial intelligence and language models.
The architecture is expected to be based on Navi 48, although there are no details about potential changes to the memory bus configuration. The RX 9070 XT uses a 256-bit bus, so to support 32 GB of GDDR6 VRAM, the memory modules would likely need to be doubled. This could require a more complex PCB design, possibly with dual-sided memory chips, a feature common in professional graphics cards.
Competition and Market Positioning

In addition to the increased memory, the Radeon RX 9000 is expected to maintain the 20 Gbps transfer rate of the RX 9070 XT. However, it would still be behind Nvidia’s RTX 50 series, which features GDDR7 memory operating at speeds between 28 and 30 Gbps.
The RX 9070 XT 32GB is expected to be released in the second quarter of 2025. The large amount of VRAM suggests that AMD might target users needing more memory for GPU-intensive applications like 3D rendering, machine learning, and high-resolution video editing.
Although there is no official word on pricing or how this model fits into the RDNA 4 series, this new GPU could become the most advanced option in the RX 9000 lineup, competing directly with Nvidia’s RTX 5090.
The addition of 32 GB of VRAM raises questions about its necessity for gaming, as 16 GB is usually sufficient for 4K gaming with high-quality graphics. However, with the growing demands of modern games and the potential use in generative AI, the larger memory could prove useful.