If I ask you how many versions of the GeForce RTX 3060 there are, you would probably tell me two, one with 12 GB of graphics memory and a 192-bit bus and another with 8 GB of graphics memory and a 128-bit bus. You are not wrong, but to these two we should add a third version using the GA104 GPU instead of the GA106 GPU.
This is a bigger difference than meets the eye, because that GPU change affects the specifications of the GeForce RTX 3060. The version based on the GA106 GPU has a configuration of 48 raster unitsalso known as ROPs, while the variant with GA104 GPU has 64 raster units.
Thanks to this improvement that the GPU incorporates, the performance of the GeForce RTX 3060 with GA104 GPU improves palpably in professional applications, and stands out especially in rendering tasks. For example, in Blender that model is up to 20% faster than the version based on the GA106 graphics core.
The rest of the specifications remain, that is, that model has 3,584 shaders and has 12 GB of 15 GHz GDDR6 memory on a 192-bit bus, which means that it will not present a significant performance difference in games compared to the model with 48 raster units . In any case, it is a relevant change that is worth keeping in mind, especially if we are thinking of buying a GeForce RTX 3060.
I remind you that the GA104 GPU is the same one that NVIDIA has used in the GeForce RTX 3060 Ti, GeForce RTX 3070 and GeForce RTX 3070 Ti, and that by having a different maximum shader configuration it can keep more raster units activated with 3,584 active shaders out of the total of 6,144 that said GPU has.
If you wonder why NVIDIA has made this GPU change, I think it is very likely that it is a simple question of stock, that is, that the ones in green are trying to clear a large number of GA104 chips that didn’t come off the wafer ready to be integrated into a GeForce RTX 3060 Ti or better. You already know that chips that do not have a specific amount of functional shaders can be reused to bring inferior solutions to life.
