Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

TPUs vs. GPUs and why Google is positioned to win AI race in the long term

uncoveralpha.com

Post reply on HN
Print view View original

271–280 of 328 posts

Re: TPUs vs. GPUs and why Google is positioned to win AI race in the long term

#271

Post by alecco »

zmmmmm wrote: post #215

I always enjoy being wrong and I was very wrong in my predictions about Google : I thought they should theoretically win, but I was also very confident they couldn't possibly turn their execution ship around to actually pull together a coherent competitor to OpenAI. But they do seem to have done that and it's very impressive. If they do continue to execute, I can't see anybody stopping them dominating and I would be…

Unless they nerf Gemini 3.0 after a few weeks like they did with 2.5. Remember?

Re: TPUs vs. GPUs and why Google is positioned to win AI race in the long term

#272

Post by socalgal2 »

zmmmmm wrote: post #215

I always enjoy being wrong and I was very wrong in my predictions about Google : I thought they should theoretically win, but I was also very confident they couldn't possibly turn their execution ship around to actually pull together a coherent competitor to OpenAI. But they do seem to have done that and it's very impressive. If they do continue to execute, I can't see anybody stopping them dominating and I would be…

It doesn’t help when their thankgiving doodle that sends me to Gemini on how to plan making thanksgiving dinner on time completely fails in ridiculous ways

Re: TPUs vs. GPUs and why Google is positioned to win AI race in the long term

#273

Post by solumunus »

thelastgallon wrote: post #62
m4r1k wrote: post #41

Google's real moat isn't the TPU silicon itself—it's not about cooling, individual performance, or hyper-specialization—but rather the massive parallel scale enabled by their OCS interconnects. To quote The Next Platform: "An Ironwood cluster linked with Google’s absolutely unique optical circuit switch interconnect can bring to bear 9,216 Ironwood TPUs with a combined 1.77 PB of HBM memory... This makes a rackscale…

Also, Google owns the entire vertical stack, which is what most people need. It can provide an entire spectrum of AI services far cheaper, at scale (and still profitable) via its cloud. Not every company needs to buy the hardware and build models, etc., etc.; what most companies need is an app store of AI offerings they can leverage. Google can offer this with a healthy profit margin, while others will eventually run…

They just need to actually make and market a good product though, and they seem to really struggle with this. Maybe on a long enough timeline their advantages will make this one inevitable.

Re: TPUs vs. GPUs and why Google is positioned to win AI race in the long term

#274

Post by mda »

bastawhiz wrote: post #186

I don't think what the article writes about matters all that much. Gemini 3 Pro is arguably not even the best model anymore, and it's _weeks_ old, and Google has far more resources than Anthropic does. If the hardware actually was the secret sauce, Google would be wiping the floor with little everyone else. But they're not. There's a few confounding problems: 1. Actually using that hardware effectively isn't easy. It…

"Gemini 3 Pro is arguably not even the best model anymore"

Arguably indeed, because I think it still is.

Re: TPUs vs. GPUs and why Google is positioned to win AI race in the long term

#275

Post by veunes »

The funniest thing about this story is that NVIDIA has essentially become a TPU company. Look at the Hopper and Blackwell architectures: Tensor Cores are taking up more space, the Transformer Engine has appeared, and NVLink has started to look like a supercomputer interconnect. Jensen Huang isn't stupid. He saw the threat of specialized ASICs and just built the ASIC inside the GPU. Now we have a GPU that is 80% matrix multiplier but still keeps CUDA compatibility. Google tried to kill the GPU, but instead forced the GPU to mutate into a TPU

Re: TPUs vs. GPUs and why Google is positioned to win AI race in the long term

#276

Post by breppp »

veunes wrote: post #275

The funniest thing about this story is that NVIDIA has essentially become a TPU company. Look at the Hopper and Blackwell architectures: Tensor Cores are taking up more space, the Transformer Engine has appeared, and NVLink has started to look like a supercomputer interconnect. Jensen Huang isn't stupid. He saw the threat of specialized ASICs and just built the ASIC inside the GPU. Now we have a GPU that is 80% matri…

There's an issue with building a swiss knife chip that supports everything back to the 80s, it works great until it doesn't (Intel)

Re: TPUs vs. GPUs and why Google is positioned to win AI race in the long term

#278

Post by Jensson »

nostrademons wrote: post #229
sidibe wrote: post #128

Earlier quoted context omitted.

They've actually had many very successful projects that make the few products and acquisitions you are thinking of work. It's true most of their end products don't work or get abandoned but it stretches their infrastructure in ways that works out well in the long run

I should probably have said "products" rather than "projects". There's a fair bit of extremely good engineering that goes on in the infrastructure side, but when it comes to consumer products, if one of the founders isn't explicitly sponsoring it it gets killed.

They get killed since ROI of putting more people on search or ads is almost always higher than ROI on the new projects, not because the founders doesn't like them.

Re: TPUs vs. GPUs and why Google is positioned to win AI race in the long term

#279

Post by veunes »

sharpy wrote: post #133
morkalork wrote: post #118

Earlier quoted context omitted.

Anti-moat like commoditizing the compliment?

If they get things like PyTorch to work well without carinng what hardware it is running on, it erodes Nvidia's CUDA moat. Nvidia's chips are excellent, without doubt, but their real moat is the ecosystem around CUDA.

The problem is that "hardware-agnostic PyTorch" is a myth, much like Java's "write once, run anywhere". At the high level (API), the code looks the same, but as soon as you start optimizing for performance, you inevitably drop down to CUDA. As long as researchers are writing their new algorithms in CUDA because it's the de facto language of science, Google will forever be playing catch-up, having to port these algorithms to XLA. An ecosystem is, after all, people and their habits, not just libraries.

Re: TPUs vs. GPUs and why Google is positioned to win AI race in the long term

#280

Post by veunes »

m4r1k wrote: post #41

Google's real moat isn't the TPU silicon itself—it's not about cooling, individual performance, or hyper-specialization—but rather the massive parallel scale enabled by their OCS interconnects. To quote The Next Platform: "An Ironwood cluster linked with Google’s absolutely unique optical circuit switch interconnect can bring to bear 9,216 Ironwood TPUs with a combined 1.77 PB of HBM memory... This makes a rackscale…

OCS is indeed an engineering marvel, but look at NVIDIA's NVL72. They took a different path: instead of flexible optics, they used the brute force of copper, turning an entire rack into one giant GPU with unified memory. Google is solving the scale-out problem, while NVIDIA is solving the scale-up problem. For LLM training tasks, where communication is the bottleneck, NVIDIA's approach with NVLink might actually prove even more efficient than Google's optical routing.
Post reply on HN