Patel explains NVLink connects only 72 GPUs while Google ICI connects 8,000 chips without switches, creating architectural trade-offs.
“NVIDIA, the NVLink can only connect 72 GPUs. For Google, their ICI can connect 8,000 chips at super high bandwidth, but you have to pass through other chips to get there because there's no switch.”
Patel reveals Google runs three separate TPU design programs with Broadcom, MediaTek, and undisclosed partner.
“They're making a TPU with Broadcom, that's a different architecture than a TPU with MediaTek, that's a different TPU than the architecture that is, I won't disclose, by research.”
Patel predicts Google and OpenAI will both release new models in two weeks.
“Release New release in two weeks. Everyone's releasing in two weeks. Google, OpenAI, maybe Anthropic. I don't know about Anthropic, but Google and open AI definitely releasing in two weeks.”
Patel notes YouTube holds unique video data that gives Google training advantage despite scraping restrictions.
“I don't know what ungodly percentage of video is only on available on YouTube. And sure, you can scrape YouTube, blah blah blah.”
Patel reports Google is backing out of Scale AI, spending around $250 million this year but reducing future commitment.
“Google's backing out, I think they're gonna spend on the order I've heard like $250,000,000 this year with them. And they're backing out.”
Patel reports Google is spending $250 million with Scale AI this year but backing out of the partnership.
“And they're backing out. Obviously, they've spent a lot of money and there's stuff they can't back out of, but it's like, that's gonna go down a lot.”
Patel says Google built rack-scale AI systems with Broadcom in 2018 before NVIDIA deployed similar architecture.
“Google did something very similar in 2018, right, with the TPU. Now they couldn't do it alone. Right? They know the software.”
Patel says Google deployed water cooling years before NVIDIA and achieves higher reliability than NVIDIA GPUs.
“Google's brought in water cooling for years. Right? NVIDIA only just realized they needed water cooling on this generation. And Google's brought in a level of reliability that NVIDIA GPUs don't have.”
Patel claims Google is massively ramping TPU production, a closely guarded secret most DeepMind employees don't know.
“Google's actually ramping up TPU production massively. And I think people in AI would be like, Well, duh. But like, okay, who has the capability of figuring out the number?”
Patel says Google's TPU production numbers are closely guarded secrets that even most Google DeepMind employees don't know.
“Right? That's like a very closely guarded secret, most people that work at Google DeepMind don't even know number.”
Patel claims Google will have more compute than any other company by a large factor.
“Google is going to have more compute than any other company in the world period by like a large, large factor.”
Patel says OpenAI has a lot less total FLOPS than Google despite having more GPUs than most perceive.
“whole point was that Google you know, OpenAI, who everyone would be like, oh, yeah. They have more GPUs than anyone else. Right? But they have a lot less flops than Google.”
post Patel quotes Deepmind friend saying TPU ecosystem grows stronger every time Googlers leave
https://x.com/dylan522p/status/2085203412040434079
post Patel asks if Jeff Dean leaving is positive or negative for Broadcom
https://x.com/dylan522p/status/2085196853373718752