Nvidia GeForce RTX 50 series Blackwell GPUs to be manufactured on TSMC's 4NP node
Kopite7kimi, a leaker with an impeccable track record, revealed some information about the GB202, GB203 and GB205 GPUs powering the next-gen GeForce RTX 5090, RTX 5080 and RTX 5070. Shortly after Nvidia showed off its flagship GB100 GPU based on the Blackwell architecture, Kopite7kimi confirmed that the GeForce RTX 50 series would be manufactured on the same TSMC 4NP (not to be confused with N4P) node.
The leaker adds that the 4NP node should result in a 30% increase in transistor density. This is remarkably higher than the 6% transistor density increase between N4 and N4P. It makes sense for Nvidia to use an enhanced version of its custom node (4N) instead of jostling for capacity on cutting-edge alternatives like N3B and N3E. Whether or not Nvidia can keep die sizes within the EUV reticle limit (856 mm2) without compromising yields remains to be seen.
Kopite7kimi adds that each SM (Streaming Multiprocessor) of GB202 (and other Blackwell GPUs) might get a larger L1 cache. This could be part of an overall SM redesign, which the leaker says will improve each SM's throughput. With the entire GB3202 die slated to feature 192 SMs, the GeForce RTX 5090 might actually deliver upon the rumoured 2.6 performance increase.