This information is quite interesting, especially considering the relatively short time since the last graphics card was released. So far there has been no word on any conversion of Ada Lovelace GPUs for use in other models, but that will soon change. According to noted insider @kopite7kimi, Nvidia plans to use the AD103 GPU for the RTX 4070 card.
To the best of our knowledge, there is no card on the market yet that takes full advantage of the AD103 GPU. The closest is the RTX 4080 with 9728 CUDA cores and only 5% of the GPU turned off. The RTX 4070 has 5888 CUDA cores, so if NVIDIA decides to use the AD103 GPU, it means that 42% of the GPU will be partially disabled.
Partners of the company should have no problems with the exchange of the GPU as the AD103 and AD104 are compatible in terms of contacts. The AD103 GPU will be significantly chopped and will have a lower TDP level, so they can easily use the same heatsinks. Questions might arise about the L3 cache size, number of ROPs, and coding capabilities, as these parameters usually affect GPU replacement.
However, it should be noted that GeForce desktop cards do not hesitate to combine different characteristics of the GPU. These cards are usually sold in the Chinese market, and there is already enough information about RTX 3070 and RTX 3060 models with non-standard GPUs. This might explain why NVIDIA does not display the exact names of the GPUs on its official website.
It’s worth noting that the use of higher GPUs in lower-level cards is not uncommon, but it usually happens later in the production of graphics cards when there are enough GPUs in stock to use. Another option is to use defective processors that have non-functional blocks. In this case, it is too early to market these cards. This raises questions. But maybe we simply don’t know everything.
