The new generation of graphics cards NVIDIA is however not expected before, at best, September 2022.
Remember, a few weeks ago, NVIDIA was the victim of an attack that allowed the theft of a lot of data. Today, this data is coming to the surface, much to the chagrin of the manufacturer.
96 MB L2 cache on the AD102 GPU?
Indeed, and even if it is necessary to take the necessary distance from all this “information”, the theft of data organized by the Lapsus$ group today makes it possible to decipher multiple information related to the future range of graphics cards of NVIDIA, the famous Lovelaces.
The five supposed references of this new range are to be put in front of the five references of the Ampere range. From the outset, it is of course the manufacturing process that distinguishes the two generations: the Samsung 8N on Ampere and the TSMC 4N for the future Lovelace.
Beyond that, we note above all a very large increase in the amount of onboard L2 cache: we go from 6 MB to 96 MB on the most muscular models… as if to counter theInfinity Cache from AMD. The number of Cuda cores also increases markedly and the same is actually true for all compartments of the GPU.
There is, however, a notable exception, in that NVIDIA seems to carry over the same memory interface buses from one generation to the next. The 128-bit entry-level models will still have a little more difficulty convincing us.
On the same subject :
Intel Arc Alchemist: the first equipped laptops available … in South Korea
Source : Semi-Analysis
6