Original title: TSMC 4N 5nm mighty! 4080 12GB transistors are far more than 3090Ti but half smaller
The NVIDIA Ada Lovelace RTX 40 series uses a special TSMC 4N process, a customized version of NVIDIA, which actually belongs to the TSMC 5nm process family.
In fact,TSMC 5nm family version is very rich, including standard version N5, performance enhanced version N5P, N4, N4P, N4X, NVIDIA 4N.
Although the naming is very confusing, the TSMC 4N process is still good in terms of specifications, far better than the Samsung N8 8nm of the RTX 30 series.
AnandTech learned the exact area and number of transistors of the first three cores of the RTX 40 series, as follows:
－ RTX 4090 AD102: 608 square millimeters, 76.3 billion transistors, 125.5 million per square millimeter
－ RTX 4080 16GB AD103: 378.6 square millimeters, 45.9 billion transistors, 121.1 million/square millimeter
－ RTX 4080 12GB AD104: 294.5 square millimeters, 35.8 billion transistors, 121.1 million per square millimeter
The three cores integrate more than 120 million transistors per square millimeter, and AD103 and AD104 are exactly the same, far exceeding the 45.1 million TSMC 8nm process RTX 3090/Ti GA102 cores.
Even more astonishingly,As a small core, AD104 has 7.5 billion more transistors than the large core of GA102, but the area is less than half of the latter, which is reduced by as much as 53%!
In addition, the Ada architecture achieves an increase in the number of ROP raster units, three cores respectively92, 112, 80which can greatly improve the performance of traditional raster games.
The second level cache is also increased to96MB、64MB、48MBwhich is expected to significantly increase the bandwidth.
Editor:
Statement: The opinions of this article only represent the author himself, Sohu is an information publishing platform, and Sohu only provides information storage space services.
