Prohardver Article: Multi-Sheet Experimental Design Was Designed By NVIDIA To Scale Machine Learning

The IT industry has been struggling for some time with the slowdown of Moore’s law. Although the statement itself could still be true today, the cost of newer manufacturing technologies is very high, and the benefits of many physics laws mean that benefits are getting smaller and it is no longer economically viable to keep Moore’s law alive; at least in the traditional way. Nevertheless, companies are typically resourceful, so if a problem cannot be solved, they try to bypass it.

NVIDIA at this year’s GTC presented an experimental multi-sheet design, which is actually a chip-like concept. Although the prototype RC18 is quite simple, it contains RISC-V modules on a 16 nm node, specifically 36 pieces. The prototype is not necessarily testing specific chiplets, but rather checking the connection.

 

To read more, please visit: https://prohardver.hu/hir/nvidia_rc18_tobblapkas_kiserleti_dizajn.html. Please note that the original article is in Hungarian.

Tags: