NVIDIA’s “beasts” gets 18176 CUDA cores

Rumored specs of the flagship RTX 40 GPU might have just been revealed by Kopite7kimi.

First and foremost, the leaker claims that AD102 based model would feature a new board design ( PG137 as opposed to PG139 for RTX 4090). There has been rumors that NVIDIA is developing a triple-fan design for their RTX 40 series, so this rumored flagship model could very well be the first card that needs such a cooler and possibly a new PCB.

And this new cooler will surely be needed, if the card really has a TDP of 800W as mentioned by Kopite. That’s significantly more than RTX 4090’s 450W TDP. Such power requirement might force NVIDIA’s to launch their first reference design with dual 16-pin power connectors.

This SKU allegedly features 142 Streaming Multiprocessors, so a total of 18176 CUDA cores . That’s around 11% more than RTX 4090. It is also said that this RTX 40 GPU would feature as much as 48GB of GDDR6X memory. It is not only higher capacity than RTX 4090 (24GB) but also faster memory (24Gbps vs 21 Gbps). Such configuration enables a maximum bandwidth of 1.1 TB/s .

Worth noting, that the leaker is not putting a name on this card. One simply cannot be sure if this is the new RTX 4090 Ti or the new TITAN model. In fact, there is no guarantee that such a model will ever be released, but given Kopite’s track record, there is certainty that such a SKU is indeed being developed.

Kopite also claims that GeForce RTX 4090 reveal is soon. Current rumors put RTX 4090 launch date as early as October, but NVIDIA tends to teaser or announce their product sooner. The only question at this point is how much sooner will Ada architecture be shown.

Source: @kopite7kimi