Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

60 billion is obviously a metric ton, but 10 billion is not that ridiculous for an SoC to clear; there are snapdragons at higher transistor counts. The AMD 5950X clears 19 billion, and it is just a CPU with no GPU or integrated RAM. I’ve got to guess the M1 transistor count is inflated a fair bit by RAM.

I suppose it’s not likely we’ll know the actual price of the M1, but it would suffice to say it’s probably a fair bit less than the full laptop.



Not sure if these numbers can believed, but apparently a 300mm wafer at the 5nm node costs about $17000 for TSMC to process

https://www.tomshardware.com/news/tsmcs-wafer-prices-reveale...

Since the M1 is 120 mm2 and a 300mm wafer is about 70695 mm2, you could theoretically fit 589 M1 chips on a wafer.

Subtract the chips lost at the edge and other flaws, you might be able to get 500 M1 off a wafer? (I know nothing about what would be a reasonable yield but I'm pretty sure a chip won't work if part of it is missing)

Anyways, that would be $17000/500, or $34 per M1 chip - based just on area and ignoring other processing costs.


The M1 Max is slightly off-square and 432mm^2. 19.1 x 22.6 seems like a decent fit.

TSMC has stated that N7 defect density was 0.09 a year ago or so. They have also since stated that N5 defect density was lower than N7.

Let's plug that in here https://caly-technologies.com/die-yield-calculator/

If we go with a defect density of 0.07, that's 94 good dies and 32 harvest dies. At 0.08, it's 90 and 36 respectively.

If we put that at 120 dies per wafer and $17,000 per wafer, that's just $141 per chip. That's probably WAAAYY less than they are paying for the i9 chips in their 2019 Macbook Pros.

For comparison, AMD's 6900 GPU die is 519mm^2 and Nvidia's 3080 is 628mm^2. Both are massively more expensive to produce.


thanks for the link to the die yield calculator, very handy!

And I agree - Apple making their own chips is sure looking good for their bottom line vs buying from Intel




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: