• AiNews.com
  • Posts
  • Cerebras Launches AI Inference Tool to Compete with Nvidia

Cerebras Launches AI Inference Tool to Compete with Nvidia

Author

Alicia Shapiro
August 29, 2024 • Estimated Reading Time: 3 minutes

An image representing Cerebras Systems' launch of an AI inference tool designed to compete with Nvidia. The central focus is a large, futuristic chip resembling the size of a dinner plate, symbolizing Cerebras' Wafer Scale Engine. The chip is shown processing vast amounts of data, represented by digital streams and data visualizations. In the background, a cloud-based environment and a sleek, modern tech interface convey the idea of scalable AI performance. The Cerebras logo is subtly integrated, emphasizing the company's innovation in AI technology. The color scheme uses modern, cool tones like blues and grays

Image Source: ChatGPT-4o

Cerebras Launches AI Inference Tool to Compete with Nvidia

Cerebras Systems, a Sunnyvale-based startup, has launched a new tool for AI developers, providing access to its powerful chips for running applications. The company claims its tool offers a more affordable alternative to the industry-standard Nvidia processors, which are widely used for training and deploying large AI models. Nvidia GPUs, often accessed through cloud computing providers, are known for their high cost and limited availability, making AI development expensive and challenging.

Unmatched Performance at a Lower Cost

Cerebras CEO Andrew Feldman highlighted the advantages of the new tool in a recent interview with Reuters, stating, "We're delivering performance that cannot be achieved by a GPU. We're doing it at the highest accuracy, and we're offering it at the lowest price." The AI inference market, expected to grow rapidly, could be worth tens of billions of dollars as AI tools gain traction among consumers and businesses. Cerebras aims to capitalize on this opportunity by offering a competitive alternative to Nvidia's dominance.

Innovative Wafer Scale Engine Chips

Cerebras' key innovation lies in its Wafer Scale Engine chips, each the size of a dinner plate. These chips address a significant challenge in AI data processing: the inability of large models to fit on a single chip, which typically requires hundreds or thousands of smaller chips to work together. Cerebras' chips allow for faster processing by eliminating this bottleneck, enabling AI applications to run more efficiently.

Flexible Deployment Options

Cerebras plans to offer several types of its inference product through a developer key and its cloud platform. For customers who prefer to manage their own data centers, the company will also sell its AI systems directly. The pricing model is designed to be competitive, with charges as low as 10 cents per million tokens—a common metric for measuring AI model output.

Preparing for a Public Offering

In a strategic move, Cerebras is preparing to go public, having filed a confidential prospectus with the Securities and Exchange Commission earlier this month. This step signals the company’s confidence in its market position and its readiness to challenge established players like Nvidia in the AI hardware space.