Nvidia’s A100 GPU was showcased on February 9, 2023 at Nvidia’s headquarters in Santa Clara, California.
Katie Tarasoff
Nvidia shares soared to a market cap of nearly $1 trillion in after-hours trading on Wednesday. After the company announced a surprisingly strong outlook for the future, CEO Jensen Huang said the company would have a “huge record year”.
The increase in sales is due to soaring demand for Nvidia’s graphics processors (GPUs), which power AI applications from Google, Microsoft, OpenAI, and others.
Fueled by demand for AI chips in data centers, Nvidia is on track to post $11 billion in revenue this quarter, beating analyst estimates of $7.15 billion.
“The ignition point was generative AI,” Huang said in an interview with CNBC. “We know CPU scaling is slowing down, accelerated he knows computing is the way forward, and a killer app has emerged.”
Nvidia believes there is a definite change in the way computers are made that could lead to further growth. Data center components could even be a trillion-dollar market, Huang said.
Historically, the most important part of a computer or server has been the central processor, or CPU. That market was dominated by Intel, with AMD as its biggest rival.
Graphics processors (GPUs) have taken center stage with the advent of AI applications that require massive amounts of computing power, with state-of-the-art systems using as many as eight GPUs per CPU. Nvidia currently dominates the AI GPU market.
“In the past, data centers were mainly CPUs for file searches, but in the future they will be generative data,” says Huang. “You’re not going to get data, you’re going to get some data, but most of the data has to be generated using AI.”
“In other words, instead of millions of CPUs, far fewer CPUs will be used, but they will be connected to millions of GPUs,” Huang continued.
For example, Nvidia’s proprietary DGX system is essentially an AI computer for training built into a single box, using eight of Nvidia’s high-end H100 GPUs and just two CPUs.
Google’s A3 supercomputer combines eight H100 GPUs with one high-end Xeon processor from Intel.
That’s one reason why Nvidia’s data center business grew 14% in the first calendar quarter, while AMD’s data center division remained flat and Intel’s AI and data center division declined 39%.
Additionally, Nvidia’s GPUs tend to be more expensive than many central processors. Intel’s latest generation of his Xeon CPUs can cost as much as $17,000 at list price. On the secondary market, one of his Nvidia H100 may sell for his $40,000.
NVIDIA will face increased competition as the AI chip market heats up. AMD has a competitive GPU business, especially in gaming, and Intel also has its own line of GPUs. Startups are developing new kinds of AI-focused chips, mobile-focused companies like Qualcomm and Apple continue to push the technology forward, and someday giant servers will be in his pocket instead of his farm. You may be able to run it inside. Google and Amazon are designing their own AI chips.
But for today’s companies building applications like ChatGPT, Nvidia’s high-end GPUs are still the chips of choice, as they are expensive to process and train terabytes of data, and generate data using models. It makes text, images, or predictions that are also expensive to run later in a process called “inference”.
Analysts say NVIDIA remains the leader in AI chips thanks to proprietary software that makes it easy to use all GPU hardware capabilities for AI applications.
Huang said Wednesday that replicating the company’s software is not easy.
“You have to design all the software, all the libraries, all the algorithms, integrate them into a framework and optimize them for the architecture, the architecture of the entire data center, not just one chip. ,” Huang said. he said on a conference call with analysts.
