During this past weekend, multiple leaks and rumours surrounding the RDNA 3-based AMD Navi 31 GPU and Nvidia's AD102 GPU have emerged. The most surprising claim is the amount of Infinity Cache that AMD's next flagship GPU will pack, with reports claiming that it will ship with 512MB.
According to the leaks and rumours summarised by 3DCenter, the Navi 31 GPU will feature a 256-bit memory interface, but it may boast 512MB of Infinity Cache to compensate for that. The GPU itself is expected to have an MCM (multi-chip module) design and come with 15,360 streaming processors divided into 2x CGDs, each with 30x RDNA Workgroup Processors (WGP).
One interesting change in design mentioned by Bondrewd is the absence of Compute Units from the Navi 31 GPU. In previous AMD GPU architectures, each WGP would contain multiple CUs.
As for Nvidia AD102 GPU rumours, the leaker @greymon55 said the Lovelace GPU will use “improved GDDR6X memory”. It's unclear what “improved” precisely means, but we assume it will be better efficiency, higher clock speeds, or both. Another leaker also stated that AD102 won't feature any sort of Infinity Cache-like memory.
Discuss on our Facebook page, HERE.
KitGuru says: We are most likely more than a year away from seeing these GPUs release, but the rumour mill is starting to paint a good picture of each of them. Considering how far we still are from getting them into our hands, specifications may still change considerably.