Nvidia's RTX 5090 will undoubtedly be a powerful GPU, partly due to the massive GB202 chip that powers it. According to MEGAsizeGPU, the chip has a size of 744 square millimeters. The chip size of RTX 5090 will significantly increase by 22% compared to the AD102 GPU of RTX 4090. This may push up prices - both for Nvidia and for end consumers.
It is rumored that the Nvidia Blackwell RTX 50 series desktop GPU will adopt TSMC's 4NP (5nm) process, which is an enhanced version of the 4N process used in the Ada Lovelace architecture (RTX 40 series). This process node is expected to bring a 30% density increase, but for the RTX 50 series, this data may not be entirely accurate. Anyway, the chip size of 744 square millimeters is approaching the mask size limit (approximately 850 square millimeters).
To overcome this limitation, Nvidia adopted a multi chip design in its data center Blackwell B200 chip, but considering factors such as high packaging costs, it is clearly not feasible for desktop products.
According to sources, the GB202 GPU size of RTX 5090 is 744 square millimeters (24 mm x 31 mm), while the overall package size is 3528 square millimeters (64 mm x 56 mm). The packaging size is not that important because they include the chip itself as well as capacitors, resistors, and other structural materials. Anyway, the RTX 5090 will be the largest chip size consumer GPU on the market since 2018. According to this leaked information, Turing still maintains its dominant position as the TU102 (RTX 2080 Ti) GPU is only 10 square millimeters larger than GB202.
Nvidia always builds some redundant parts in its GPU, so the top-level RTX 5090 solution may not have fully enabled the chip. Normally, fully enabled "Prime" chips are reserved for use in data centers and professional GPUs. It is reported that the fully enabled GB202 chip provides 192 SM units, and according to current leaked information, the RTX 5090 has 88% enabled GB202 chips (170 SM units).
This is completely consistent with the RTX 4090, which is also limited to 88% (128 out of 144 SM units). According to reports, the RTX 5080 has been severely weakened, with only 84 SM units, so its chip size may be about half that of GB202. Nvidia is likely to bear high manufacturing costs for the RTX 5090, but this will undoubtedly be passed on to the end consumers. The price difference between RTX 5080 and 5090 Blackwell graphics cards is expected to be significant.
Recently, leaked information has revealed that Huang Renxun will make his debut on Nvidia's flagship RTX 5090 and RTX 5080 GPUs at CES 2025. Subsequently, mainstream RTX 5070 and 5070 Ti GPUs are expected to be launched later in the first quarter, followed by budget RTX 5060 products.
