During the presentation of GeForce Beyond, NVIDIA paid the most attention to the flagship AD102 GPU. But about the AD103 and AD104 located one step lower, which can be found in the GeForce RTX 4080 16GB and RTX 4080 12GB, respectively, so little has been said that not a word at all. After a couple of days, the chipmaker corrected this misunderstanding.
Against the backdrop of a truly monstrous AD102, which received an almost record 76.3 billion transistors, the younger chips, although they look impressive, are not so impressive. The AD103 has an area of 378.6 mm² and 45.9 billion transistors, while the AD104 has an area of 294.5 mm² with 35.8 billion transistors. For comparison, the GA102 GPU in the RTX 3090 Ti has an area of \u200b\u200b628 mm2 and “only” 28.3 billion transistors.
And since the RTX 4080 12GB is actually the RTX 4070, which became “4080” solely to justify such a high price, it’s safe to say that the RTX 4070 received a more capacious chip than the previous generation flagship.
The lion’s share of the transistor budget fell on the additional L2 cache. In the case of the AD103 chip, this is 64 MB – an increase of 16 times relative to its predecessor. The block diagram of the AD102 core clearly shows that the L2 cache occupies about 25% of the chip area.
Finally, NVIDIA is misleading about the process technology. It would seem, what can be meant by TSMC N4, specified in the specifications of the cards? In fact, this is a 5-nm process technology with modifications, it does not meet the 4-nm standards.