Hacker News Reader

Comment by Octoth0rpe | Hacker News Reader

> We're all just waiting for Intel to give up on the Arc GPUs because of it.

I'm sure all kinds of radical cost cutting is being proposed internally, but I'd be surprised if this was on the line. My thinking is that Arc GPUs are the "walk before you run" product that will lead to Intel producing AI accelerators, which is a reasonable bet re: a pathway to profitability.

View on HN

A lot of projects are "walk before you run". Didn't stop Intel from killing them and wasting all their time, money and effort. That's Intel's problem. I don't want them to kill Arc. It's what they need long-term. I simply don't believe they have the persistence and long-term vision to keep at it despite lackluster sales and lack of interest.

"AI accelerators" are at peak profitability right now. Within a few years, certainly by 2030, chip optimizations and improvements will see Raspberry Pis or your phone doing AI work.

I expect we may even aee AI cores on phones, by 2036.

Some phones have TPUs on them today...

Phones are already getting NPUs / TPUs / Neural Engines / whatever fancy marketing term they can come up with.

I think software optimizations are going to heavily impact gpu/ai-accelerator demand. It feels like we’re leaving the ‘brute force’ make it work era and entering the ‘make it work on my phone without killing battery life” era. I wouldn’t be shocked if within a few years there just isn’t the need for custom hardware at all.

Reply on news.ycombinator.com