The GeForce RTX 40 will mark a leap as big as the GTX 10

GeForce RTX 40

We have been quite busy for a few days with numerous rumors dedicated to the GeForce RTX 40, the next generation of graphics cards from NVIDIA for the general consumer market, whose launch will occur, if all goes according to plan, between the end of 2022 and the beginning of 2023, Although most sources strongly point to the last quarter of 2022 as the date of the debut of that new generation.

Everything seems to indicate that NVIDIA will use two different architectures for its next products; Ada Lovelace, which will be the basis for the GeForce RTX 40, and Hopper, that it will be the first NVIDIA architecture to use an MCM-like design, and that it will be limited to its professional products. It makes sense, at the end of the day it would not be the first time that the green giant uses different architectures for the consumer and professional sectors, it already did it with Pascal and Volta, but it is not confirmed, so we must continue to take that information with extreme caution.

We know that Ada Lovelace will maintain the monolithic core design, and that it will be manufactured in the process of 5 nm from TSMC. Some details have also emerged about the possible specifications of the AD102 graphics core, a chip that will be used, in theory, to give life to the GeForce RTX 4090, and that will feature 18,432 shaders, 144 RT cores (if NVIDIA keeps Ampere base, i.e. one RT core for each SM drive), 576 tensor cores (provided that the same distribution of 4 texture units is maintained for each SM unit seen in Ampere), 576 texturing units and 128 raster units.

It is evident that, in terms of performance, the jump would be enormous just because of the difference in specifications that we can see compared to the RTX 3090, a graphics solution that mounts 10,496 shaders. However, we must not forget that, beyond the raw power of the increase in SM units, the GeForce RTX 40 will also offer higher performance thanks to the improvements of its new architecture, Ada Lovelace.

GeForce RTX 40: A huge generational leap?

The latest rumors point in that direction, which assure that the Ada Lovelace architecture will mark such a great advance like the one we saw at the time with the passage from Maxwell to Pascal. To better understand what this means, it is necessary to make some comparisons with equivalences between graphics cards of different generations.

When NVIDIA launched the Pascal architecture, there was an even bigger leap than we experienced at the time with the evolution from Kepler to Maxwell, and it was palpable in both performance and efficiency. Thus, the GTX 980 Ti, the most powerful of NVIDIA at that time, was overtaken by the GTX 1070, and the GTX 980 was almost on the same level as the 6GB GTX 1060.

For comparison purposes, remember that the GTX 980 Ti had 2,816 shaders, 176 texturing units, 96 raster units, a 384-bit bus and a TDP of 250 watts, and it was surpassed, as we have said, by the GTX 10’70, a graphics card of only 1,920 shaders, 120 texturing units , 64 raster units and a 256-bit bus, whose TDP was also 100 watts lower (150 watts).

Personally, I don’t think the GeForce RTX 40 achieves the same leap in efficiency (performance per watt consumed) that we saw with Pascal, but they are likely to mark a similar advance in terms of gross power. If this is confirmed, the performance equivalences within the GeForce RTX 40 series could be distributed in a similar way to what we saw with the arrival of the GTX 10, which means that:

  • The GeForce RTX 4090 would have no equivalent.
  • The GeForce RTX 4080 would be much more powerful than the RTX 3090.
  • The GeForce RTX 4070 would offer similar performance to the RTX 3080 Ti.
  • The GeForce RTX 4060 would perform around the same level, or slightly less, than the RTX 3080.
  • The GeForce RTX 4050 could perform similarly to an RTX 3060.

Keep in mind that this estimate that I have offered you is based on our own analysis, which I have made based on all the information we have and my knowledge within the sector. It makes sense, but it might not be true.

Like it? Share with your friends!


Your email address will not be published. Required fields are marked *