Nvidia GeForce RTX 50 Series: Rumored Use of TSMC’s 4NP Node

Nvidia GeForce RTX 50 Series Boosted Transistor Density and Enhanced Node Utilization

Nvidia-centric leaker, Kopite7kimi, has divulged that the GB202 GPU, which powers the GeForce RTX 5090, will be fabricated on TSMC’s 4NP node. This is the same node used for the GB100 GPU. The leaker also hinted at an increase in throughput for each Streaming Multiprocessor (SM).

With a flawless leak history, Kopite7kimi has shared insights about the GB202, GB203, and GB205 GPUs. These are set to power the forthcoming GeForce RTX 5090, RTX 5080, and RTX 5070. Following Nvidia’s showcase of its flagship GB100 GPU, founded on the Blackwell architecture, Kopite7kimi confirmed that the GeForce RTX 50 series would also be manufactured using the TSMC 4NP node, similar to the GB100.

GB202 will use the same process node as GB100. I must clarify once again that TSMC 4N(vidia) is based on TSMC 5, not 4nm.
I'm sorry I cannot match Jensen's naming with TSMC's naming. We need professional chip analysis to determine.
At least, there is a 30% increase in density.

— kopite7kimi (@kopite7kimi) March 19, 2024

The leaker further noted that the 4NP node should lead to a 30% surge in transistor density, a significant leap compared to the 6% increase between N4 and N4P. It’s a logical move for Nvidia to utilize an enhanced version of its custom node (4N) rather than competing for capacity on advanced alternatives such as N3B and N3E.

Nvidia RTX 50 Series Potential Performance Increase and SM Redesign

However, whether Nvidia can maintain die sizes within the EUV reticle limit (856 mm2) without affecting yields is yet to be determined.

Kopite7kimi also suggested that each SM of GB202 (and other Blackwell GPUs) could receive a larger L1 cache. This might be a component of an overall SM redesign, which according to the leaker, will boost each SM’s throughput. Considering the entire GB3202 die is expected to feature 192 SMs, the GeForce RTX 5090 could potentially fulfill the rumored 2.6 performance increase.