The development of genAI technology, the future of artificial intelligence, raises questions about its impact on data centers in the coming years. It is clear that the application of generative artificial intelligence (generative AI) and other machine learning methods will […]
The development of genAI technology, the future of artificial intelligence, raises questions about its impact on data centers in the coming years. It is clear that the application of generative artificial intelligence (generative AI) and other machine learning methods will be widespread across all industries. However, training AI models with trillions of parameters and trillions of data is incredibly expensive, and finding an efficient way to execute these AI processes is crucial.
The largest genAI models require tens of thousands of graphics processing units (GPUs) for training over a period of three to four months. To execute inference, which involves generating responses, a server with eight GPUs is required. If GPUs do not become smaller and cheaper, CPU manufacturers will need to enhance the mathematical capabilities of their devices to retain a portion of this work. However, it remains uncertain whether CPUs will be powerful enough to assume the role of GPUs. If CPUs fail to provide sufficient power for mathematical operations, there is a risk that genAI will not be cost-effective for widespread adoption.
Intel currently has a weak position regarding mathematical operation accelerators, and its neural processors have not been fully adopted for genAI inference. If Intel fails to create more advanced and affordable GPUs compared to Nvidia and AMD, the company’s CPUs could lose the race in data centers. As artificial intelligence becomes increasingly prevalent in business environments, it is unclear how long the distinctions between AI and other workloads will exist, and whether CPUs or GPUs will dominate these data centers.
Pat Gelsinger, Intel’s CEO, has emphasized the need for further development in the field of data centers to maintain the company’s competitive position. Intel plans to capture a larger portion of the accelerator market in 2024, with the help of its AI accelerators like Gaudi. Greater growth is expected for Intel in the coming years, but competition from AMD and Nvidia will pose significant challenges.
The total revenue from AI training and inference in 2023 could reach approximately $50 billion. Intel aims to profit from this market and currently has an opportunity worth around $2 billion. However, it is important to note that the pipeline, or the estimate of potential jobs, always exceeds the actual revenue.
What is generative artificial intelligence?
Generative artificial intelligence is an area of artificial intelligence that focuses on developing models capable of creating original data, such as images, texts, or sounds.
What are the major challenges in implementing genAI technology?
The major challenges in implementing genAI technology are the high costs of training AI models, the need for significant computational power, and efficient real-time execution of inference.
What factors should be considered when examining data centers in the context of genAI technology?
When considering data centers in the context of genAI technology, it is important to take into account the performance and price of GPUs and CPUs, as well as their ability to perform the mathematical operations required for training and inference of AI models. Monitoring competition among manufacturers, such as Intel, Nvidia, and AMD, is also crucial to assess their future dominance in this field.