Hopper will be the name of the NVIDIA architecture that replaces Ampere and could be based on chiplets

Rumor indicates that NVIDIA is already working on the replacement of Ampere, which will be called Hopper and could be based on a chiplet system

Currently on the market are NVIDIA Turing graphics cards, which feature RT Cores for ray tracing. Rumors suggest that the Ampere series , based on 7nm lithography and with improved ray tracing capabilities, will be released in 2020. This is everything we know about the next generation of graphics cards. Now, a new name, Hopper, has emerged as a potential replacement for Turing.

As with Pascal, Turing, and Ampere, the name is not chosen randomly. Grace Hopper is a very important person in the computer world. She was an eminent American military and computer scientist. He developed the first linkers and programmed the Harvard Mark I computer, the first electromechanical computer. This computer was used by the United States during World War II.

Hopper will be Ampere's replacement in the future

Who has revealed this information is 'Kopite7Kimi', a Twitter user who has leaked many things from NVIDIA. Normally he throws the tweet and then deletes it. On this occasion, he said that his next architecture will be called Hopper and will have a major change. Hopper will have a different graphic design based on a chiplet format.

AMD has already bet on this design with Zen with remarkable success. Intel proposes in future solutions to do something similar with Foveros, which will be stacked processors. A few months ago it was rumored that AMD could bring the chiplet to graphics, although in the end it has not been the case.

NVIDIA would implement it in Hopper, although we don't know much on what it would be based on. AMD uses Infinity Fabric which is heavily dependent on RAM frequency. Intel will use EMIB in Foveros, a more efficient system that does not depend on other elements. NVIDIA may have its own technology or use third-party technology.

Using this system will allow highly scalable graphics with a huge number of cores. You could even create three silicons: one filled with CUDA Cores, one filled with RT Cores, and one filled with Tensor Cores. For gaming it may not seem relevant, but for solutions like the Tesla or Quadro, it could be a bombshell.

We reiterate, this is a rumor, the company has not said anything about this.

Source: TPU