Hacker News Reader

Comment by BearOso | Hacker News Reader

Of course. When you buy out all the capacity of the TSMC 3nm node for a year and use it to make huge, low-yield chips and charge a small fortune for them, you can get some performance advantages.

Intel's fabs are catching up, though, and have more money invested in them again. Ideally, we'll see a more competitive market soon.

View on HN

>use it to make huge, low-yield chips and charge a small fortune for them, you can get some performance advantages.

1. It is not huge, by CPU Core or by SoC Die Space.

2. It is not low-yield either.

3. Comparatively speaking, what Apple include their SoC as BOM / COGS isn't anywhere as high as Qualcomm charges, hardly a small fortune.

4. It is performing close but at a much lower wattage. That is not just coming from a node advancement but mostly microarchitecture.

Do you have a source for these claims?

Literally every industry publication and BOM analysis since... the first iPhone?

Which one? All of these are nearly publicly available information or can be directly inferred. Except they are either from Annual Report, Investing Meetings, Court Documents, or actual Geekbench numbers. The problem is people read Macrumours or WCCftech and believe they have informed.

Another factor is that Apple designs big cores exclusively for consumer apps on iPhones. There is no compromise for server workloads, high clock targets, chips with modest die areas, slow memory, dies without efficiency cores, AI SIMD and so on.

I don't think any other company in the world can financially justify such a specialized big core, except maybe Samsung (who has failed in their ventures so far).

Do they? It seems their system architecture between the A-series for phones and the M-series for their pads/laptops/etc is largely the same. It's certainly the same design facility.

The applications they target are very similar.

Editing 8k video and running a social media app on a phone are vastly different workloads.

> Editing 8k video

I would hope the GPU would be handling this.

...But honestly, they are not that different. Both are heavy media processing, and you would build the same kind of CPU core for either as opposed to (for instance) a database workload.

Video encoding if you need the best quality still is done on the CPU.

Video encoding also favors Apple's wide, SMT free design.

In theory chunked encoding gives you the best quality, and the best CPU for that is a bunch of little e-cores, but the only app I've really seen implement chunked encoding as a "max quality" feature is av1an.

But editing 8k video in a social media app on a phone is a similar workload. And really, TikTok’ers are doing stuff like that with their phones these days.

Reply on news.ycombinator.com