Yesterday AMD officially released Radeon RX9070XT and Radeon RX9070, which are based on the new generation of RDNA4 architecture GPU and support AMD FidelityFX SuperResolution4 (FSR4) technology. Both new graphics cards are equipped with Navi48 chips, and both have 16GB of GDDR6 video memory.
In fact, before the official release, there was news that the Navi48 chip has 53.9 billion transistors and is approximately 350mm² in size, which is smaller than the previously rumored 390mm². Judging from the data finally announced by AMD, the area is slightly larger at 357mm², and the number of transistors is exactly the same.
In other words, the size of Navi48 is smaller than NVIDIA's AD103 (378.6mm²) and GB203 (378mm²), but the transistor density is higher. The transistor density of Navi48 reaches 150MTr/mm², while AD103 and GB203 are 121.2MTr/mm² and 120.6MTr/mm² respectively. AMD is about 25% higher than this generation of NVIDIA Blackwell architecture GPUs. This should be one of the GPUs with the highest transistor density so far. Considering that AMD and Nvidia use similar manufacturing processes, this gap is estimated to exceed many people's expectations.
NVIDIA does not seem to care so much about transistor density in the current generation of products, and the Blackwell architecture is even lower than the previous AdaLovelace architecture. Since transistor counts are often considered approximate and there are different counting methods, and the results are taken from official specifications, the actual difference may not be that big.
It is actually a bit difficult or unfair to compare the transistor density increase between RDNA4 and RDNA3. The former is a single-chip design, while the latter is a small chip design combined with MCM packaging. GCD and MCD use 5nm and 6nm processes respectively. This time AMD returns to a single chip without sacrificing density or efficiency. Combined with improvements in architecture and technology, it seems to have good prospects.