Can Ibm’s New Chip Dethrone The Industry Leader Nvidia?

By Author: joy
Total Articles: 177
Comment this article

The race to develop and deploy generative artificial intelligence (GenAI) has proven to be a costly endeavor, with OpenAI's operational expenses for ChatGPT hitting eye-popping figures. The reported daily operational cost of running ChatGPT alone has reached a staggering $700,000, and the company spent a remarkable $4.6 million over two weeks to train its GPT-3 model. These astronomical costs stem from several factors, including the number and price of Graphics Processing Units (GPUs) required for training and running these advanced AI systems.

GPUs play a pivotal role in the development of GenAI technologies. OpenAI utilized a staggering 9,200 GPUs to train its GPT-3 model, underscoring the immense computational power needed for such undertakings. However, the expenses don't end at the initial purchase of GPUs; they also encompass the significant power usage associated with these demanding computations. Presently, Nvidia dominates the market for chips used in GenAI, with their H100 Tensor Core GPU and A100 Tensor Core GPU reigning supreme. This dominance has prompted Nvidia to make substantial investments in its product, ...
... banking on its market leadership.

Yet, the landscape might be on the brink of change, with IBM's announcement of a potentially groundbreaking chip with exceptional energy efficiency. IBM unveiled a prototype 14nm analog chip that exhibits the potential to be up to 14 times more energy-efficient per watt compared to current GPUs. While currently a research project, this chip holds the promise of delivering reduced energy consumption and cost efficiencies for enterprises engaged in GenAI projects, such as GPT-4 and Midjourney.

The uniqueness of IBM's chip lies in its analog nature. Unlike conventional digital chips that work with binary signals (1s and 0s), IBM's analog chips can comprehend gradations between these values by manipulating analog signals. The incorporation of phase-change memory in the chip allows it to switch between amorphous and crystalline phases in a manner akin to the binary operations of digital processors, as well as intermediate states between these values. This analog chip is capable of performing computations directly within memory due to its compute-in-memory components.

IBM has expressed that these 14nm chips have the potential to encode a remarkable 35 million phase-change memory devices per component and can model up to 17 million parameters. The implications of this advancement are substantial. Notably, one of the most significant challenges faced in AI today revolves around the energy and time expended while transferring substantial amounts of data between a processor and data storage. This energy consumption can range from 3 to a staggering 10,000 times the energy expended for the actual computation. IBM's analog chips have the potential to alleviate this challenge by significantly reducing energy usage, costs, and environmental impact.

The potential benefits for AI-centric businesses are substantial. The deployment of IBM's analog chips could revolutionize the AI industry by offering energy efficiency, cost savings, and reduced environmental impact. Furthermore, these chips could reshape the trajectory of AI development and provide enhanced computational power, thereby influencing the course of GenAI advancement.

IBM's experiments with its 14nm chip have yielded positive results, including highly accurate audio transcriptions of spoken content. However, the true impact of these analog chips on the GenAI landscape hinges on whether IBM decides to mass-produce them. If this development gains traction, it could potentially redefine the future of GenAI chips, offering a more energy-efficient and environmentally conscious solution to the resource-intensive demands of AI computation. As the field continues to evolve, IBM's analog chip innovation has the potential to set a new benchmark for energy-efficient AI technology.

Read More: https://www.techdogs.com/tech-news/td-newsdesk/can-ibms-new-chip-dethrone-the-industry-leader-nvidia