Hacker News Reader

Comment by thisIsFiner | Hacker News Reader

Apple Silicon launched in Mac, the first couple iterations worked/sold great. By the launch of M3 market realized it wasn't a fluke and that x86 was dead but still fundamental due to deployment at the time

Flash forward a couple more years to now it's all about efficient model training which sucks on generic CPUs so everyone is buying and designing other chips

Two big consumer pivots over 4-5 years and boom ...they're IBM and handed lifelines to shore up chip manufacturing in the US for strategic reasons but Intel essentially manufactures rotary phones right now. Maybe the fab biz will get modern designs right. We'll see

View on HN

Modern Intel chips have no issues trading blows with Apple Silicon.

Their fabs are the weakest part of their business (I wonder why once AMD spun off GlobalFoundries so they could focus on the core product, they suddenly started doing well) Once they started making chips on TSMC's processes, wow, all of a sudden, theyre in the same league as Apple Silicon.

Intel should have been split in two many years ago. The fab side and the chip design services side. Together they are dragging each other down.

Aren't they separate ?

Actually, Intel's fab's are becoming their crown jewel. The whole "5 nodes in 4 years," where they tried to catch up with TSMC pretty much worked, which is why the stock originally rocketed up by 5x a few months ago. They aren't far behind TSMC now.

But, they still don't have many external customers for the fabs. Intels fab revenu is something like a tenth of TSMC's. Although, it'll probably grow year by year if they continue to match the capabilities of TSMC's chip processes.

Maybe no issues "trading blows" in a single reductive compute per second measure

But Intel chips use more electricity to do so. At data center scale that is not something shrugged off like homelab nerds shrug off the extra $50/month or whatever on their power bill.

> At data center scale that is not something shrugged off

Sure it is. At datacenter scale, there are precisely zero HPC vendors that have invested in Apple Silicon as an alternative to Ampere or Grace CPUs. Same goes for datacenters seeking to replace their Intel CPUs. Chips like the M-series are explicitly nonviable for Xeon-scale use cases.

We are not discussing the same things and I will leave it at that.

Feel free to write a full response when you get the time.

Reply on news.ycombinator.com