6 points | by meetpateltech 5 hours ago
3 comments
> we expect Maia 300 to deliver 40% more tokens per megawatt compared to leading GPU-based systems.
Apparently, not as performant as Cerebras or Groq?
a day late, and a billion dollars short.
Of course they were building their own chips - to reduce their dependence on Nvidia, just as was predicted years ago. [0]
That Terafab is predicting a scenario where a single company is going to be used as a chokepoint in Taiwan.
So guess who the customers of that will be when that happens?
[0] https://news.ycombinator.com/item?id=39750169
> we expect Maia 300 to deliver 40% more tokens per megawatt compared to leading GPU-based systems.
Apparently, not as performant as Cerebras or Groq?
a day late, and a billion dollars short.
Of course they were building their own chips - to reduce their dependence on Nvidia, just as was predicted years ago. [0]
That Terafab is predicting a scenario where a single company is going to be used as a chokepoint in Taiwan.
So guess who the customers of that will be when that happens?
[0] https://news.ycombinator.com/item?id=39750169