Hacker News Reader

It’s kind of strange their benchmark for “Machine Learning Programmers” is “Simulation of Dynamical Systems in MATLAB.” Seems like they could have capitalized better by using a generative ai inference benchmark.

View on HN

iPhone 15 Pro NPU can do 35 TOPS. M3 can only do 18 TOPS. That's why they didn't focus on ML in the video.

Its big question mark with 35 TOPS, because its extremly limited and hard to achive.

Maybe they know M3 is not up for that kind of task.

Possibly this, but also consider Apple is playing major catch up with generative AI. I could see avoiding mention just to keep it a bit further from the mind of reviewers.

Could be, but people are already using M2 Ultra for LLM stuff and like it. Seems like Apple would double down on that.

I agree the machines have been good to run various ML models that aren’t running on other PCs in their category. (Though, shrinking the resource requirements seems like a never ending exercise.)

Apple likes to wait until it can announce something with a most around it. In this case, more compute isn’t enough, I don’t think.

Reply on news.ycombinator.com