Nvidia Engineer Sharply Responds to Intel Head’s Claim of Luck in AI Dominance
An employee, formerly with Intel, asserted that the “Green Team’s” superiority stems from vision and execution, not luck.
A few days ago, Intel CEO Pat Gelsinger caused widespread bewilderment by attributing Nvidia’s current AI market dominance to sheer luck. He claimed that if Intel hadn’t discontinued its supercomputing GPU project in the 2000s, Nvidia would have been in trouble. This project was shut down just as Intel “let go” of Gelsinger, so he might have personal grievances about the situation. An engineer, who worked on Intel’s failed “Larrabee” project and now at Nvidia, countered Gelsinger, arguing that it wasn’t about “luck” but Nvidia’s strategic vision and effective execution – qualities Intel lacks.
The outspoken Nvidia employee is Brian Catanzaro, noted in his biography as the Vice President of Applied Deep Learning Research at Nvidia. He previously appeared in videos about Nvidia’s Deep Learning Super Sampling (DLSS) technology. In his tweet, he mentioned working on Larrabee in 2007, and his LinkedIn profile shows he was an intern at Intel then.
In 2012, Intel claimed Larrabee could perform ray tracing in modern games at an acceptable frame rate, suggesting the GPU was ahead of its time. There are parallels between Intel’s ambitions for Larrabee in 2007 and what Nvidia introduced in 2018 with the Turing architecture. (Possibly, Catanzaro significantly contributed to implementing this technology in Nvidia’s GPUs, sharing Intel’s developments after leaving the company.)
At that time, Nvidia and Intel were fiercely competing in high-performance computing: Nvidia was pushing its Tesla GPUs, and Intel its GPGPU Knights Ferry. Journalists noted, “Seeing modest interest in Nvidia’s Tesla platform in high-performance computing applications, Intel hoped its compatibility would maintain Intel’s dominance in supercomputing technologies.”
The situation turned out unfavorably for Intel, as it essentially abandoned the Larrabee project, choosing instead to use its x86 processors for HPC tasks and focusing its graphics efforts on integrated GPUs. When Intel officially announced Larrabee in 2010, AnandTech wrote that Intel’s decision “fully validated Nvidia’s Tesla strategy.”
Catanzaro noted in his tweet that back then, Intel was ten times larger than Nvidia. Hence, the company believed Larrabee would crush Nvidia. However, it lacked the vision and execution to advance the project to what Nvidia is now doing with its GPUs in parallel computing. A quick Google search confirms these figures: in 2007, Nvidia’s annual revenue was about $3 billion, compared to Intel’s $38 billion.
Fast forward to the third quarter of 2023, and the figures are vastly different. Nvidia earned $18 billion in the quarter, 206% more than the previous year, while Intel made $14 billion, an 8% decrease. More importantly, Nvidia earned $14.5 billion from its data center products, while Intel made just $3.8 billion in this area.
It remains to be seen whether this verbal spat will continue. It’s unlikely to last long, as Gelsinger probably wouldn’t want to appear disrespectful towards one of Intel’s key partners. In previous statements, he mentioned his friendship with Nvidia’s CEO and their frequent interactions, raising the question of whether this topic ever came up in their discussions.