For months we have been immersed in everything that has to do with the new AMD and NVIDIA graphics cards, of which we learn day by day their characteristics or their performance. Weighed how soon a year will be filtered the possible yields of the Porcentuales on the actual RTX 30 y RX 6000 of the back companies, esos datos fueron borrado y ahora han vuelto actualizados, donde hay sorpresas muy interesantes y que vuelven a sacar a la luz fantasmas the past. What performance they will have the new graphics from AMD and NVIDIA?
There are two parts in this news, the first as we say is already old and which sets the base, and the second which gives new information on what leaked at the time. For this reason and given that a year is a long time, we are going to update the information and above all put it in context, because since we have no other choice, we will have to enter the field of the speculation we’re so good at after seeing the successes.
The performance of new AMD and NVIDIA graphics cards
What we saw back then was that NVIDIA and its AD102 for its RTX 40 would mean a performance jump of more than double, specifically 2.2X, over what was the fastest GPU at the time and which is still on sale, the RTX 3090. As for AMD, its jump would be even greater (2.5X) and it would be, if not at the height of its eternal rival, ahead and by the minimum.
11 months ago I posted a tweet about performance estimation of AD102(2.2x), Navi31(2.5x) and GH202(>3x).
Then I deleted that quickly, because I didn’t want to start a war.
– kopite7kimi (@kopite7kimi) April 29, 2022
Finally, we had Hopper as an architecture, which was to be launched with a GH202 chip, which in the end is not the c ase and the GH100 with Hopper micro architecture was in charge of said performance (3X).
How much truth is there in these statements? Well, since we haven’t seen anything on performance, we can’t verify whether the chips are that far above their predecessors as such, since we only know the specs of one of them. and that it actually points to a similar figure thanks to its 18,432 Shaders and 144 SMs. But there is more…
NVIDIA changes to Ada Lovelace
Y es que el propio leaker weighs a lanzar de nuevo el tweet donde vuelve a especificar los mismos datos alegando a una guerra que no sabemos a quién afectaría, ahora también determined que aquel AD102 que nombró no es el actual AD102 que NVIDIA ya tiene terminado y What It’s not just a soda amp as advertised at the time.
This means that, as expected, NVIDIA will introduce significant changes (notably in the performance of the RT Cores and Tensor Cores) that will improve the data offered at this time.
I should clarify that the current AD102 is NOT the original AD102. Ada Lovelace is no longer a mere refresh of Ampere, even though it was that way in the beginning.
– kopite7kimi (@kopite7kimi) April 29, 2022
The conclusions we can draw are simple: NVIDIA’s leap forward will be greater than that 2.2X and may end up matching AMD’s leap (2.5X), so if it was the ‘expected gap, the tables would turn They would continue with distances similar to those between the current RTX 30 and RX 6000.
If these speculations come true, the battle will be tough again, except that on this occasion and apparently there will be no price war because they will be high and therefore the margin of movement will be greater, especially since NVIDIA starts from a node that is exclusive: 4N, like the one used by the GH100. What we are sure of is that the performance of new AMD and NVIDIA graphics will not take one, but two steps forward in a single generation.
