Breaking Moore's Law: Nvidia's AI Chips Redefine Tech Progress

[Renegade]

Nvidia CEO Jensen Huang claims that Nvidia AI chips are outpacing Moore's Law as of now, something he likes to call the "hyper Moore’s Law". The Nvidia's GB200 NVL72 chip is reported to be 30-40x faster than its predecessor for AI inference tasks. Nvidia is able to sustain this growth by developing architecture, chips, and software simultaneously, something that Apple claimed with Apple Silicon too. Expect rapid performance improvements across training and inference stages as chips become more efficient.

This rapid progress will lead to AI costs getting cut down drastically, making advanced models like Open AI o3 cheaper and easily accessible. As inference costs keep going down, AI will integrate deeper into common use devices. Future chips may power models with smarter reasoning and improved learning efficiency, creating a feedback loop of better AI at lower costs. I still feel that all this breakneck speed of improvement and AI becoming mainstream will continue for next several years before it starts to slow down. But who knows, maybe we will finally make a breakthrough in quantum computing soon.

Source: https://techcrunch.com/2025/01/07/nvidia-ceo-says-his-ai-chips-are-improving-faster-than-moores-law/

 

[aadilxdev]

That a good thing because when the AGI paper was out the amount it took for open ai o3 was soo much costly, and this would also open option on hosting inhouse open source models cheaper with larger parameters
But since its just available for their flagship model for now it seems we have to wait until we get something like this for consumer gpus too

 

[aryabhattadey]

View attachment 219701Nvidia CEO Jensen Huang claims that Nvidia AI chips are outpacing Moore's Law as of now, something he likes to call the "hyper Moore’s Law". The Nvidia's GB200 NVL72 chip is reported to be 30-40x faster than its predecessor for AI inference tasks. Nvidia is able to sustain this growth by developing architecture, chips, and software simultaneously, something that Apple claimed with Apple Silicon too. Expect rapid performance improvements across training and inference stages as chips become more efficient.

This rapid progress will lead to AI costs getting cut down drastically, making advanced models like Open AI o3 cheaper and easily accessible. As inference costs keep going down, AI will integrate deeper into common use devices. Future chips may power models with smarter reasoning and improved learning efficiency, creating a feedback loop of better AI at lower costs. I still feel that all this breakneck speed of improvement and AI becoming mainstream will continue for next several years before it starts to slow down. But who knows, maybe we will finally make a breakthrough in quantum computing soon.

Source: https://techcrunch.com/2025/01/07/nvidia-ceo-says-his-ai-chips-are-improving-faster-than-moores-law/

He will obviously try to hype up his chips. But any independent sources on this beyond what he said?