An interesting rumor ensures that NVIDIA will use the graphics core GA102 to create a GeForce RTX 3090, a graphics card that could be the top of the range for the next generation of the green giant. I say “could” because there is always the possibility that they decide to accompany it with a GeForce RTX 3090 Ti.
This information makes sense, a lot of sense. On the one hand it would allow NVIDIA to better differentiate its top-of-the-range models from those that are located in the high-end, and on the other hand it would be a return to the strategy they adopted in previous generations. No, it would not be the first time that the X800 and X900 are differentiated, remember, for example, what happened with the GeForce FX 5800 and GeForce FX 5900, or with GeForce 7800 and GeForce 7900.
If we look at more current generations, the last to follow this approach was the GTX 600 series, with the GTX 680 and GTX 690, although the second was a dual GPU graphics solution that departs from the classic concept.
GeForce RTX 3090: 4,992 shaders and 12GB of GDDR6 memory
The source of the rumor ensures that this new graphics card will have 4,992 shaders, a figure that may seem a little strange but that would be perfectly viable, since with the current structure of 64 shaders for each active SM we would have a total of 78 SMs. If NVIDIA does not introduce changes in the division of its other elements and maintains the Turing values, the rest of its specifications would be configured such that:
GA102 graphic core.
312 texturing units.
98 raster units.
78 RT cores for ray tracing.
624 tensor cores for artificial intelligence.
12 GB of GDDR6 memory at 16 GHz-18 GHz.
Keep in mind that the specifications of that supposed GeForce RTX 3090 that we have listed above are an estimate calculated on the basis of the proportions of SMs, shaders and others that Turing uses. If NVIDIA changes the 64 shader rule for each SM the values would be totally different.
Regarding performance, I remind you that according to the latest information, this graphics card could be 30% to 40% more powerful than the RTX 2080 Ti. We talk about performance in gross terms, that is, in general. If we limit ourselves to ray tracing the performance improvement could be much bigger and around 50%.
The presentation of the new NVIDIA graphics architecture is expected for the second half of 2020, but the launch of the new GeForce RTX 30 series will not take place. until the end of the year.