The other observation is that the precision will almost always be higher in the accumulation step than in the multiplication step. This is specific to AI chips. You're multiplying low-precision numbers, and then when you accumulate, errors accumulate quickly, so you need more precision there.
You might also like
There's actually been startling progress in reinventing the fundamental nature of the car itself — something the industry calls the 'software-defined vehicle,' controlled by just a handful of powerful computers instead of dozens or even hundreds of independent electronic control units, or ECUs. Xinzhou says that moment is basically here.
Nilay Patel — Decoder with Nilay Patel
The thing that I find most exciting is that transformers have essentially won. We know that this is the architecture. And if you know that, then you can build a chip that only runs transformers, and you can get massive efficiency gains by doing that, because you're not building a general purpose chip.
Invest Like the Best with Patrick O'Shaughnessy
When I first started in 2016, I really think beauty was about expression and now increasingly it's about optimization. So people have kind of moved from like how do I want to be? How do I want to express myself to how do I fix myself? How do I optimize?
It's Been a Minute