The rumour comes by way of popular leakster, kopite7kimi (@kopite7kimi), who tweeted that the NVIDIA graphics card with the Lovelace-based GPU architecture will feature a new board design – PG137, as opposed to PG139 – and that the reference card for the GPU may just feature a new triple-fan design. How the latter will be executed, however, remains to be seen, seeing how an earlier leak and rumour suggests that the GPU brand may be planning on recycling the cooler design introduced with the Founders Edition of its high-end GeForce RTX 3080 and RTX 3090. Getting back on point, kopite also stated that their source says that the top-tier Lovelace card will sport an AD102-450-A1 GPU, which is further complemented by 48GB of GDDR6X running at 24Gbps. Not only that, but kopite says that the top-tier card will also house as many as 18176 CUDA cores, which is approximately 11% more than what is believed to be the number of CUDA cores of the supposed RTX 4090.
— kopite7kimi (@kopite7kimi) July 25, 2022 However, the kicker to the rumour about the NVIDIA GPU is that the top-tier card will require a TDP of 800W, in order to run efficiently. That is significantly higher than the initial 450W TDP speculation, as well as double the amount of GDDR6X graphics memory and much faster memory. If these rumours are to be believed, then the alleged Lovelace GPU could very well have a maximum bandwidth of 1.1TB/s. Once again, this is just a rumour and as with all news of such nature, you’d do best to take it with the usual application of skepticism. In this case, Kopite doesn’t even specify the exact Lovelace GPU bearing these god-tier specifications, but given NVIDIA’s habit of reserving the best for its most premium and enthusiast-level cards, it could very well be the much anticipated RTX 4090 Ti. (Source: Videocardz)
