A little background, this announcement comes out of Nividia's yearly GPU Technology Conference, so we can all expect to see additional news pieces coming out over the next hours and days.

This supercomputer is being used in part to demonstrate the capabilities of Nivida's line of commercially available supercomputers such as the DGX Pod and DGX Superpod, these are expensive AI supercomputers which are being marketed to wealthy corporations, government agencies, and the exorbitantly rich who can afford them for their private labs.

For context, when the more expensive version of the DGX line, The Superpod, was created it made the Top500 list for most powerful supercomputers, ranking 22nd at the time. Back in August, Nvidia created the 7th most powerful supercomputer, Selene using Superpods, probably in preparation for this, but that was all with the old A100's. This new design using H100's is much faster and clearly DESIGNED specifically for training large language models. This fact cannot be emphasized enough, even if 100T doesn't result in proto-AGI's or anything of that sort, it becomes increasingly clear that Nvida's plan seems to be mass producing models and using them to build the Metaverse. If Nvidia is even moderately successful in doing this, along with the other AI companies, it will result in an information explosion comparable to the birth of the internet, it would be an internet with constant autogenerated content, AI capable of, at minimum, speeding up the development of all media and code-based resources, video games, film, music videos, and proprietary software.

Something like this happening prior to recursive improvement is what we might call a soft takeoff in which progress greatly accelerates prior to the creation of true AGI/ASI but improves enough to begin increasingly changing the world around us for a relatively lengthy period (perhaps years) before culminating in the creation of more fully conscious systems.

A major emphasis of this conference has been the Omniverse (Nvidia's brand for the Metaverse). Since the Metaverse concept started picking up in business several months ago and Nvidia revealed several new developments at that time back in November. Nivida has been branding themselves as the provider of infrastructure for that omniverse, not only helping build the supercomputers they believe will power it, but also the "connective tissue".

For this reason it is no doubt, a major point of pride for Nvidia, to make heavy usage of their Quantum 2 Infiniti Band here, which in very simple terms, you could think of as extremely fast fiber optic cables which have been touted as being a major part of the future infrastructure for the metaverse...the fundamental idea here being that the future involves powerful AI supercomputers connected using cutting edge networks like this...a sort of new, more powerful version of the internet built using extremely fast connections which Nvidia would like to get contracted to lay the cable for.

Just one observation about compute, eventually all this compute will become accessible on a more consumer level based on how this stuff gets less expensive, so eventually we'll see speeds like 70 TB/per sec everywhere and it will change everything.

Analyses of the likes of Nvidia, Meta, Google, and OpenAI has not been taken seriously enough, these companies are poised to change the world. It becomes too easy to assume what they're doing is hype, but their intentions and opinions have been revealed, sometimes not so subtly through their actions and statements.


45추 받은 댓글이네