Η AMD tries to implement the use of a smaller channel bandwidth in conjunction with a new technology that hears the name "infinity cache".
Κas we approach the release date of the news AMD RDNA2 GPUs codenamed “Big Ships", More and more information and rumors are appearing, causing discussions in the online forums.
We have already mentioned in the information that the flagship of the new GPUs from AMD will be RX 6900. The card will probably come with 16GB GDDR6 VRAM in one 256-bit bandwidth. Rumors of using HBM memory are unlikely to be substantiated, as using such memory would be costly.
Instead, AMD is trying to implement the use of a smaller channel bandwidth in conjunction with a new technology called "infinity cache".
The user momomo_us located on the site Justia Trademarks reference to this new technology and relayed the relevant information.
From VideoCardz we read that internally AMD does not use the name "Infinity Cache", but it seems that the company changed its mind recently. Everyone is wondering what exactly this technology is achieving, but for now it is a mystery.
The most common assumptions are that it may be an L1 GPU cache shared between cores or some sort of cache interface throughout the GPU. Certainly the name is reminiscent of "Infinity Fabric", the proprietary interface technology used by AMD for its Ryzen processor subsystems.
Is it possible for this technology to be able to balance the smaller memory bus bandwidth of the Ampere, but at a lower cost than using GDDR6X or HBM memory? If so, then AMD will have a significant advantage over Nvidia, at least in terms of pricing its cards.