Nvidia’s current generation of GPUs uses the Turing architecture, and the next generation will turn to Ampere. But according to the latest revelations, the company will continue to roll out hopper’s architecture, or after Ampere. The new architecture, named after the French physicist Andr?-Marie Ampere Ampere, is expected to be released sometime in 2020. He will be replaced by “Hopper”, named after Grace Hopper, an American computer scientist.
(Instagram via TechReport)
During World War II, Hopper is known to have programmed the Mark I computer and created the first machine language compiler team. COBOL, which was followed by COBOL, became one of the earliest and longest-running advanced computer programming languages.
Hopper GPUs don’t just change their name, they introduce a multi-chip module (MCM) package, according to twitter user kopite7kimi. It is worth noting that Nvidia has always focused on the study of MCM architecture.
To overcome the theoretical upper limit of single-chip GPUs(transistors are difficult to shrink indefinitely), the MCM architecture has emerged. It breaks down work into less complex small GPU modules (GPMs), making it easier to manufacture and increase capacity, reduce waste and improve costs.
In a paper published in June 2017, Nvidia proposed an MCM GPU design that is said to improve the performance of real-world manufacturable GPUs by 45.5%. In addition, with the same streaming processor and memory bandwidth, the performance of the GPU scenario is 27% faster.