Full Specifications for NVIDIA AD102, AD103, and AD104 GPUs

The green giant has confirmed the full specifications of its three new generation graphics cores, the NVIDIA AD102, a top-of-the-range chip that will be used in the GeForce RTX 4090; the AD103, which will be used in the GeForce RTX 4080 16 GB; and the AD104, which will be used in the GeForce RTX 4080 12 GB. Please note that these GPUs may also be used in other graphics cards coming later, such as the hypothetical (but likely) GeForce RTX 4090 Ti, GeForce RTX 4080 Ti, and GeForce RTX 4070.

With the Ada Lovelace architecture at NVIDIA, they have introduced important changes at a technological level that have made the GeForce RTX 40 the most advanced graphics cards that exist today, but we also find deep modifications in each of those graphical cores that go beyond process switching and increasing the shader count. NVIDIA has greatly increased the number of transistors, and has also clarified that TSMC 4N node is not 4nm but actually 5nm. We have updated all of our recently published articles to reflect this.

The jump to the 5nm node, and those changes at the architecture level, have allowed the green giant raise maximum shaders to nothing more and nothing less than 18,432 units in the NVIDIA AD102 core. In the AD103 we have a maximum count of 10,240 shaders, and in the AD104 we find 7,680 shaders. These numbers allow us to confirm that only the GeForce RTX 4080 12 GB uses a full graphics core, and therefore the NVIDIA AD102 and AD103 may be used to create Ti versions of the models presented, that is, a GeForce RTX 4090 Ti and a GeForce RTX 4080 Ti, respectively.

NVIDIA AD102 Graphics Core Specifications

  • Ada Lovelace architecture, manufactured at TSMC’s 5nm node.
  • 76.3 billion transistors.
  • Transistor density: 125.5 million.
  • 144 SM units.
  • 18.432 shaders.
  • 576 textured units.
  • 576 tensor cores.
  • 144 núcleos RT.
  • 192 raster units.
  • 96 MB of cache L2.

NVIDIA AD103 Graphics Core Specifications

  • Ada Lovelace architecture, manufactured at TSMC’s 5nm node.
  • 45.9 billion transistors.
  • Transistor density: 121.1 million.
  • 80 SM units.
  • 10.240 shaders.
  • 320 units of textured.
  • 320 tensor cores.
  • 80 cores RT.
  • 112 units of raster.
  • 64 MB of L2 cache.

NVIDIA AD104 Graphics Core Specifications

  • Ada Lovelace architecture, manufactured at TSMC’s 5nm node.
  • 35.8 billion transistors.
  • Transistor density: 121.1 million.
  • 60 SM units.
  • 7.680 shaders.
  • 240 units of textured.
  • 240 tensor cores.
  • 60 cores RT.
  • 80 units of raster.
  • 48 MB of L2 cache.

We can see that in addition to the increase in the density of transistors, the increase in shaders and the change in the manufacturing process, there has been with Ada Lovelace a notable increase in the total number of texturing units, and in the number of tensor and RT cores, something totally understandable as we have more SM units per GPU. The number of raster units has also been increased considerably compared to Ampere, as the NVIDIA GA102 and GA103 cores had 112 and 96 raster units, respectively.

The upload of L2 cache It is another of the most important changes that NVIDIA has introduced in Ada Lovelace, although we must not forget other details that we already told you recently when breaking down the keys to the GeForce RTX 40, such as the introduction of fourth generation tensor cores and cores 3rd generation RT.

With all this information in hand we can get an idea of ​​what is to come within the NVIDIA catalog. The GeForce RTX 4090 Ti should be based on the full NVIDIA AD102 core, and the GeForce RTX 4080 Ti will almost certainly use the full AD103 chip. Finally the AD104 will serve, in a cut version, to shape a GeForce RTX 4070.


Source: MuyComputer by www.muycomputer.com.

*The article has been translated based on the content of MuyComputer by www.muycomputer.com. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!