Today, AMD introduced the RDNA 4 graphics architecture and two graphics cards based on it — the Radeon RX 9070 XT and RX 9070. Both new products are based on different modifications of the Navi 48 graphics processor. The older model uses the Navi 48 XTX, and the younger one received the Navi 48 XT. According to the Tom’s Hardware portal, these chips have one of the highest transistor densities per square millimeter among GPUs.
Image Source: AMD
The Navi 48 chip area is 357 mm². This is smaller than the GB203 chip (375 mm²), which is used in the GeForce RTX 5080 and RTX 5070 Ti graphics cards. However, the transistor density of the new GPU from AMD is higher. In total, Navi 48 contains 53.9 billion transistors, while GB203 has 45.6 billion. Navi 48 has a transistor density of about 151 million/mm², while GB203 has 120.6 million/mm². This figure is 25% higher for the AMD chip than for its competitor. Moreover, the transistor density of Navi 48 is higher than that of the GB202 GPU (123 million/mm²), which is used in the flagship GeForce RTX 5090 graphics card.
Obviously, this isn’t a fair comparison, since the same RTX 5090 will be significantly faster than the Radeon cards unveiled today anyway. However, it’s interesting to note that Nvidia doesn’t seem to have prioritized transistor density as much when designing Ada Lovelace as AMD did for its RDNA 4 GPU. And of course, we have to keep in mind that transistor counts are generally considered approximate, and there are different ways to count them.
AMD has decided to abandon the chiplet design used for the RDNA 3 GPUs in the new generation of graphics architecture and returned to a monolithic one. This means that all the cache memory, which was previously located on separate chiplets of the graphics processor, is now contained in a single crystal along with the graphics cores. And although the transistor density of the same Navi 31 (with a chiplet design) was also about 150 million / mm², this figure was calculated without taking into account the shared use of space with cache memory on the same crystal. AMD should be given credit at least for the fact that it managed to return to a monolithic GPU design, maintain the previous level of transistor density, and also integrate 64 MB of Infinity Cache without sacrificing efficiency.