Meet RSC, Meta’s New Supercomputer

February 16, 2022 - 7 minutes read

Supercomputers are impressive machines capable of performing amazing feats of computing power. They are not a particularly new technology, with the first supercomputers coming online in the mid-1960s. Supercomputers run or support many of today’s largest companies and industries across the globe. IBM’s Watson is a famous example of a supercomputer that combines
artificial intelligence (AI) with complex analytics.

Meta announced its formation in December 2021 after years of Facebook being the company brand. A significant reason was to build the metaverse. It did not take long for Meta to announce the unveiling of the AI Research SuperCluster (RSC). RSC will lay the metaverse groundwork to expand AI, augmented reality (AR), and virtual reality (VR) technology.

So, why is this important for businesses to pay attention to? Well, we believe this will have a radical impact on application development practices through the implementation of these futuristic cloud-native computing applications, and app development platforms. So, much so that as an iPhone App Developer in Seattle, we have seen many of our clients asking questions regarding how to prepare for this major shift.

In this brief piece, we hope to answer that question and shed some light on this futuristic burgeoning technology. 

Meet Meta’s New Supercomputer

There is one primary intent for Meta’s new RSC, and that is to power their vision of the metaverse. RSC’s focus is on AI research and modeling. However, RSC will be the first piece of a more extensive AI infrastructure and framework. 

RSC’s design is a major step towards the next generation of advanced AI. Meta claims that RSC is the fifth fastest in the world, and they plan to use that computing power to train large models in natural language processing (NLP) and computer vision for research. By mid-2022, they anticipate RSC to be the fastest supercomputer in the world.

 Meta’s RSC is built using many GPUs and combining them into computer nodes. There are 760 DGX A100 nodes from NVIDIA. These nodes connect using high-performance network fabric. If this supercomputer sounds powerful, that’s because it is. Eventually, this network fabric will connect 16,000 GPUs, making it perhaps the most extensive network ever deployed.

 The creation of RSC designed to advance Meta’s AI technology and capabilities is a natural evolution. Their journey began in 2013 with the Facebook AI Research lab. When RSC reaches its full potential, it will increase the number of GPUs from 6,080 to 16,000. This significant upgrade in computing power will increase AI training by 2.5x faster. Even now, RSC runs NVIDIA’s Collective Communication Library (NCCL) nine times faster than Meta’s previous AI research clusters. 

How Will RSC Be Used?

RSC will handle new and advanced AI models that will learn from trillions of examples. These models support the end goal of building new AI systems. A more advanced AI system could perform real-time language translation faster than today’s current capabilities. Meta’s engineers hope to accomplish this goal by training AI with scaling complex and adaptable models. By doing this, Meta develops AI that is
self-supervised learning. The better AI can reason, the more it can mimic the human brain. This process helps identify patterns in massive amounts of data.

 Meta hopes that their research and training with AI will allow RSC to identify harmful content. Examples include hate speech, discrimination, or cyberbullying that plague today’s social media. The longer-term vision includes the potential for virtual meetings comprised of people worldwide. Using digital, 3D avatars, they would all speak in their native langue. RSC will translate while the participants interact in the virtual meeting room with each other.

Security and Privacy

Meta’s struggles with privacy violations are not new. Because of this reason, data security and integrity are major concerns for RSC. Meta has a treasure trove of data that pours in from its family of applications. These apps include Facebook, Instagram, and WhatsApp. According to Meta, privacy and security were a focus from the initial design for RSC. The only way to build new AI models on the complexity and scale for RSCs intent is to use real-world data. The data comes from these popular applications.

 The designers and engineers for RSC implemented end-to-end encryption for the entire data path, with the processes to always ensure this requirement. Data does not get imported to RSC before it goes through a privacy review, which includes anonymization. Before use in AI models, the data goes through an encryption process. When finished, RSC deletes the data, along with the decryption keys.

 Perhaps the most vital safeguard is that RSC does not connect to the open internet. There are no direct inbound or outbound connections. The only traffic that RSC receives comes from Meta’s production data center. While no security measure is fool-proof, isolating RSC from the internet is a powerful filter.


The metaverse comes closer to us every day, and Meta’s RSC supercomputer is a significant step closer. AI, AR, and VR are the primary technologies envisioned to engage with the metaverse. Today, plenty of applications and hardware use AR and VR. Gaming is the first industry that comes to mind. However, education and research leverage these technologies to great effect. 

 The future of mobile applications and how we engage with the internet continues to evolve. The metaverse’s excitement and potential come from how it may change our digital engagement and interaction. 

Here at Dogtown Media, we have a strong track record in mobile applications that focuses on AI, AR, and Machine Learning. Connect with us today to see how we can bring your mobile applications to market ready for the metaverse.

Tags: , , , , , , , , , , , , , , , , , , ,