This is the fastest computer in the world with which Zuckerberg wants to conquer the metaverse


Madrid

Updated:

Keep

By now everyone knows about the name change from Facebook to Meta. The new technology brand reflects the ambition of its CEO, Mark Zuckerberg, for the conquest over the next ten years of the metaverse: that kind of alternative world in which, they are convinced, we can meet friends, work or do sport through the use of Virtual and Augmented Reality technology. We are still a long way from that reality, but Zuckerberg hopes that the scientific advances of the next decade will make it possible.

To continue climbing steps, Meta has announced that it is building the fastest Artificial Intelligence computer to date, a cluster of computers – devices linked by a high-speed network – called the “AI Research SuperCluster.”

It will be finished in mid-2022 or, at least, that is what the company points out.

The objective of this SuperCluster is precisely to help Meta scientists to train new models, such as natural language or computerized vision, where the computing power determines the number of parameters that can be used in their development, in this case, these can exceed 3,000 million.

The main objective is the development of the appropriate algorithms and models for the metaverse, as well as augmented reality tools, natural language recognition that allows simultaneous translation in real time in different languages, and computerized vision.

So that we can size this new SuperCluster, we must take into account that it will be 20 times faster than the current ones used by the Meta research team. The first that the company has was launched in 2017 with 22,000 NVIDIA V100 Tensor Core GPUs. The new structure of 16,000 GPUs 20 times faster, will be able to process an exabyte of variables, or what is the same, andl Equivalent of 36,000 years of high-quality video.

The power and speed of this cluster is essential to advance in the research of new tools under self-supervised learning models that need huge databases to train.

Unlike the previous research cluster that used Meta, which only allowed the use of open source code and data for public use, this new one will allow the use of Meta code, for, for example, increase the detection of harmful content in your applications, or improve the user experience on all Meta platforms; from WhatsApp to Instagram or Facebook. This thus allows a direct use of the research results in their products.

Using private data to train these models, such as user posts on their platforms, requires a new level of security, which is why the super cluster of computers you will not be directly connected to the internet, and user data will be fully encrypted until AI training begins. Additionally, the paths from the data centers to the GPUs are fully encrypted and undergo multiple privacy and security checks in the process.

Meta wants to demonstrate its commitment to the metaverse, and the development of artificial intelligence is one of the great pillars on which this future parallel reality is based.

See them
comments


www.abc.es

Related Posts

Leave a Reply

Your email address will not be published.