I'm not going to pretend to be hyper tech-savy but just on the face of it. The WCCF Tech article doesn't add up. A 12CU dGPU just make zero sense. That's what their mainstream current gen APUs top out at. Their weakest dGPU, the 9060 (non XT), has 28 CUs and that's been their low water mark since RDNA3 (7600S laptop GPU). I know there is a segment of the internet that has a huge hate boner for MLID. I get it, the guy can be smug at times, but he gets far too many things correct to be pulling things out of his rear like the average tech article commenter would have us believe. Given the alleged IPC improvements of RDNA5/UDNA a 24, not 12, CU GPU would have raster performance similar to a 9060 with substantially better RT and PT performance. That's believable.
We'll see the truth of the matter at some point. However the leaked GPU philosophy that AMD is seemingly going for is very smart. It allows them to use the same GPU dies across multiple product segments. The memory solution, especially with the low end SKUs, provides a potential good product for the entry level market. It would finally move us beyond the ridiculous 8GB frame buffer while being cheap enough from a business standpoint to actually come to market. That said, AMD often has good ideas on paper for their GPU business. For the sake of competition, I hope it pans out this time.
