Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Exclusive: Nvidia CEO says his AI chips are improving faster than Moore's Law

Jan 08, 2025 - techcrunch.com
Nvidia CEO Jensen Huang claims that the performance of Nvidia's AI chips is advancing faster than the historical pace set by Moore's Law. During an interview with TechCrunch, Huang stated that Nvidia's latest datacenter superchip, the GB200 NVL72, is 30 to 40 times faster at running AI inference workloads than its predecessor, the H100. This advancement is attributed to Nvidia's ability to innovate across the entire stack, from architecture to algorithms. Huang argues that this rapid progress will drive down the cost of AI inference, countering concerns about the high expenses associated with AI models like OpenAI's o3, which uses a significant amount of compute during the inference phase.

Huang also introduced the concept of three active AI scaling laws: pre-training, post-training, and test-time compute, which he believes will further accelerate AI development. Despite skepticism about the slowing progress of AI, Huang remains optimistic, suggesting that Nvidia's advancements will continue to reduce costs and enhance performance. He emphasized that Nvidia's AI chips are now 1,000 times better than those from a decade ago, surpassing the traditional expectations of Moore's Law.

Key takeaways:

```html
  • Nvidia CEO Jensen Huang claims that the performance of Nvidia's AI chips is advancing faster than Moore's Law, with their latest datacenter superchip being more than 30x faster for AI inference workloads than the previous generation.
  • Huang suggests that AI progress is not slowing and introduces three active AI scaling laws: pre-training, post-training, and test-time compute, which contribute to AI model advancements.
  • There are concerns about the cost of running AI models with test-time compute, but Huang believes that more performant chips will eventually lower these costs, making AI reasoning models more affordable.
  • Huang asserts that Nvidia's AI chips today are 1,000x better than those from 10 years ago, indicating a much faster pace of advancement than Moore's Law, and he expects this trend to continue.
```
View Full Article

Comments (0)

Be the first to comment!