Performance does not squeeze toothpaste: The RTX 40 graphics card architecture may not change much

In 2022, NVIDIA will launch the RTX 40 series graphics card, and now it is basically determined that the server version of the GPU architecture is codenamed Hopper, and the game card architecture is codenamed Lovelace, which are 5nm processes.

The latest revelations say that The architecture of The Lodge has not changed much compared to Ampere and will be based on the latter. However, this does not mean that the 5nm process of the Levelace architecture GPU performance will not change, because the 5nm process upgrade will bring great changes, compared to ampere's current use of the Samsung 8nm process to improve too much, NVIDIA can greatly improve performance from the frequency and core scale.

Edited by the Institute of Core Research

Previously, it was announced that the large core number of the 5nm Lovelace architecture will be AD102, there will be as many as 12 GPC (graphics processing clusters), 72 TPC (texture processing clusters), 144 SM (streaming multiprocessors), and each SM continues to have 128 stream processors (CUDA cores), then the overall number of up to 18432 stream processors, compared to the current ampere architecture GA102 core of 10752 cores increased by 71%, Double overall performance.

Specifically, some whistleblowers have said that the frequency of The Levelace graphics card can reach 2.5GHz and the floating-point performance can reach 90TFLOPS, which is 152% higher than the RTX 3090 and 125% compared to the RTX 3090 Ti. The upgrade of Lovelace graphics cards is a bit like the original NVIDIA from 28nm Maxwell to 16nm Pascal, the architecture has not changed much, but the process improvement has still made the performance of the new generation of graphics cards soar, and the energy efficiency is very good.

(7865240)