Intel Vows to Chase NVIDIA’s AI Stride After Larrabee Slip

Recently, Intel’s CEO, Pat Gelsinger, attended an event at the Massachusetts Institute of Technology, engaging in discussions with students and participating in media interviews. During these interactions, Gelsinger addressed the trending topic of Artificial Intelligence (AI), inevitably touching upon the competition with NVIDIA. Riding the wave of AI, NVIDIA emerged as a star player in the semiconductor industry in 2023, commanding the highest market share in the critical hardware resources driving the AI revolution.

Gelsinger remarked that Intel should have been at the forefront of the AI market. However, he acknowledged the fortuitous circumstances that benefited NVIDIA’s founder and CEO, Jensen Huang. While respecting their competitor, Gelsinger did not concur with their approach. He expressed regret over Intel’s decision to abandon its Larrabee project, believing that perseverance could have brought Intel a similar fortune to NVIDIA.

In response to Gelsinger’s comments, NVIDIA’s Vice President of Applied Deep Learning Research, Bryan Catanzaro, countered on social media. He argued that Intel’s core issue was a lack of foresight and execution, making success in their previous plans unlikely.

NVIDIA initially broke into the high-performance computing (HPC) market with its “Tesla” GPU and CUDA, a unique software stack enabling developers to build and accelerate applications on their hardware, dating back to 2007. To create the CUDA ecosystem, NVIDIA invested considerable effort. They built a formidable moat with a top-down extensive software stack, including programming languages, APIs, pre-built computing, and AI models, and nurtured a vast community. Beginning with the “Volta” GPU, NVIDIA accelerated its focus on AI, realizing the need for specific hardware features to expedite the construction, training, and inference of deep learning neural networks, leading to the development of Tensor cores. As AI underwent a massive revolution in the computing field, NVIDIA was comprehensively prepared.

Intel’s Larrabee project, conceived in 2008, suffered from the company’s focus on CPU-centric chip design, relegating accelerators to a lower priority without adequate attention. Moreover, Intel’s similar projects shared a common issue: slow market introduction and a lack of ecosystem development around these chips, unlike NVIDIA’s strategy.