Hacker News new | ask | show | jobs
by voidspark 422 days ago
> They own the ecosystem thanks to CUDA, Infiniband, NGC, NVLink,

No they do not. The article explains that Google, Amazon, Microsoft, and Meta are developing their own hardware and software for AI/HPC.

Google Gemini was not trained using CUDA or Nvidia hardware.
1 comments
liuliu 422 days ago
Only one of the 4 companies you mention is successful at this. And it will remain that way.

Chinese CSPs are the only ones can develop their own hardware / software for AI / HPC.

link
voidspark 422 days ago
Wrong. It's all explained in the article.

https://www.reuters.com/technology/artificial-intelligence/m...

https://azure.microsoft.com/en-us/blog/azure-maia-for-the-er...

https://aws.amazon.com/ai/machine-learning/trainium/

link
liuliu 422 days ago
Of course corporations will have a lot of different bets. Most of them will not pan out but they will try.

Meta will not be able to produce a chip that can run GenAI workload in the next 2 years.

Microsoft is doing a side-quest, and they haven't even proved themselves with their FPGA adventure and ARM server adventure.

Amazon is legit, they have done well on ARM server, but trainium is TBD, and how much they will pull back in a recession given Jassy is a number guy will be a question mark.

No need to discuss, we can just see this in 2 years, everything will be crystal clear.

link
voidspark 422 days ago
Yeah that's right. The chip may not be as good as Nvidia's, but it doesn't need to be. As the article explains, Nvidia can still lose their position even if they have the best chips.
link
solidasparagus 422 days ago
They have to be competitive. TPUs are wildly ahead of the pack. And even they aren't particularly competitive. 12 years of ecosystem development by the most advanced AI ecosystem company on the planet and your (ex-Google!) researchers are still going to pelt you with tomatoes if you tell them you are swapping out their H100 cluster with TPUs. JAX remains niche (not saying bad) and extremely hard to use efficiently without the help of Google (no CUDA for going off the beaten path).

I suspect the closed nature of the ecosystem will preclude them from winning as much as they could.

link