Can Intel catch up with NVIDIA?
The battle to pull the rug out from beneath NVIDIA’s feet is now in full swing. NVIDIA was the phoenix that rose from obscurity in the last 10 years to blanket out the AI horizon with their GPUs.
In hindsight, they were in the right place at the right time to cash in on this AI revolution. Jensen Huang and his team have indeed been building GPUs for the gaming industry since NVIDIA’s inception. What started as a garage company for gamers decades ago now has the most high-tech chips in the most sophisticated labs on the planet.
The secret sauce for NVIDIA has undoubtedly been its CUDA playbook.
CUDA is NVIDIA’s own development platform, which makes GPU processing run on steroids. When instructions get sent to the GPU’s for processing, CUDA breaks them down into hundreds of smaller tasks to be executed at scale across multiple GPU’s. This powerhouse combination of GPUs alongside CUDA has helped NVIDIA make inroads not only into AI, or Generative AI, but into several other key areas of applications as well.
It takes more than just a platform and GPU designs to make a dent in a giant-dominated industry like chips. You have to think like a giant in order to compete with them. This is where the Omniverse, their most recent product, comes into play. They now have access to a plethora of options, particularly to make inroads into the lucrative Industrial Internet of Things (IIoT) industry.
Listed here below are a handful of them.
Digital Twins
This is their best play, in my opinion. The undeniable advantage is the capacity to leverage technology to influence results from the very beginning. There are many practical uses for digital twins. Using a digital twin to plan an entire factory, including its logistics and execution, before any work is done could be beneficial.
Indeed, a few clients are already using the Omniverse for one of their German factories (BMW for instance). The capacity to manage expenses and anticipate potential malfunctions can help prevent unnecessary shutdowns.
Factories of the future would be built by Digital Twins before a single brick is even laid. These digital factories would be run by our digital cousins, the robots. Which brings me to the next opportunity.
Robotic Applications
Since the first robot to debut on television a few decades ago, robotic applications have grown in popularity. These days, autonomous robots move objects and fill shelves on the Amazon store floor at will, all under the supervision of a human overseer.
With NVIDIA’s Omniverse offering, users can now train their robots to perform tasks such as manipulation and navigation in a simulation environment powered by Isaac Sim. They make it simpler for clients to bring these difficult jobs to market by providing a comprehensive Universal Scene Description (USD) framework for modifying robotic models and settings. Furthermore, they have the ability to generate synthetic data, which increases the usefulness of robot training.
The Metaverse
What became of this, anyway? The pipe dream or dream of Mark Zuckerberg? Despite Zuckerberg’s switch to large language models and the diminished commotion over the metaverse, action is still underway. In fact, the Omniverse launched during the peak of the Metaverse’s popularity.
Even if you were to look at the metaverse in its current, unhyped state, you’d find it in the many virtual worlds that popular games like Roblox, Minecraft, and many more have created and invaded gamers’ lives.
Creating an avatar and setting up the entire scene in a metaverse or virtual world used for business information exchange and meetings with participants wearing headsets would be difficult. It is evident that not all tech platforms can now perform these tasks; thus, in addition to handling data at scale, you would also need a platform that could edit visuals and render an entire 3-D scene in real time.
Here again, the Omniverse platform has the tools that can be leveraged to create virtual worlds and could even interact with headsets and other virtual reality devices to give users that 3D experience.
The Threat
It becomes clear why the big players are all of a sudden scrambling for a response. Not only does NVIDIA possess one of the critical hardware pieces of the future, but they are also racing ahead, making inroads in areas where others would have to sprint to catch up.
NVIDIA has amassed a war chest of its latest cutting-edge GPUs in all four public clouds (including OCI), deploying them and making them available to all users. This happened in a year when interest in AI peaked, with the LLM hype playing more in their favor, causing it to cross the USD 1 trillion market cap last year.
As of now, most of the LLM execution on the public cloud is done on NVIDIA’s GPUs (although there are other non-NVIDIA GPUs out there too).
Even as NVIDIA stretches out its tentacles into fields such as robotics and digital twins, it threatens a stranglehold on these industries, almost akin to the grip that Google has on internet search and Intel has on silicon chips. The fact that these two industry titans, along with a few others, have banded together to form a consortium of businesses aimed at neutralizing NVIDIA is truly hypocritical.
The Angle of Attack
The heart of NVIDIA’s success is CUDA, and that’s where the attack surface is. Leading the attack is a group of companies under the umbrella of the UXL Foundation. This is a grouping of companies that are collaborating with each other to develop open standards for accelerated computing on hardware.
One of their trending open source projects is OneAPI, which in itself has seven separate libraries that are on-going, including neural networks, data analytics, math kernels, threading, and a few more. Intel is throwing its weight behind this consortium, as are a lot of other chip companies such as ARM, Qualcomm, and Samsung, who are equally disgruntled at NVIDIA’s incursion into their safe space.
However, among all these other players, Intel stands the most to lose from NVIDIA’s rise and sees this alliance as a way back in.
Last year, in December, at its AI Everywhere conference, Intel announced, among other things, that its latest Gaudi accelerator (which it claims can outperform NVIDIA in GPU performance) would be out early in 2024.
This alliance has yet to gain any traction, and right now, with a 1.5 trillion market cap, NVIDIA is soaring ahead.
