Rumor: Nvidia GeForce RTX 40 flagship gets 18,176 CUDA cores and 800W tbp

Nvidia leaker kopite7kimi has published specifications of the possible GeForce RTX 40 flagship. The video card would, among other things, get a tbp of 800W. It is not known when and under which product name this alleged GPU will be released.

The possible RTX 40 flagship will receive an AD102-450 GPU with 142 streaming multiprocessors, for a total of 18,176 CUDA cores, claims kopite7kimi. This leaker more often shares correct information about Nvidia video cards early; for example, he shared months before release in 2020 the correct specs of Nvidia’s RTX 30 lineup.

The RTX 40 flagship would also have 48GB of GDDR6X memory with a speed of 24Gbit/s. According to the leaker, the total board power, or how much power the entire video card consumes, is around 800W. To illustrate: Nvidia’s current top model, the GeForce RTX 3090 Ti, has a tbp of 450W and 24GB of 21Gbit/s memory.

The card may also receive a new board design, under the name PG137. Given the memory arrangement and the high tbp, it is obvious that Nvidia will introduce a new pcb design and a new reference cooler. According to previous rumors, the GPU designer is working on a reference card with three fans.

Kopite7kimi previously shared possible specifications of the GeForce RTX 4090, RTX 4080 and RTX 4070. The flagship video card that the leaker now mentions would be positioned above those three video cards. For example, the GPU has more cores and memory and a higher tbp than the RTX 4090, which should appear later this year. It is not known under which product name Nvidia will release the flagship video card. The leaker himself does not name the card. It may be an RTX 4090 Ti or a new Titan video card.