Hacker News Reader

Comment by metamet | Hacker News Reader

Learning about how they product bin CPUs blew my mind years ago.

iirc, the CPUs come off the same manufacturing line--but the ones they sold as 3.2ghz vs 3.4ghz were based on stability tests. It was the same chip, just that some came out the oven better than others. So they would cap it and slap a label on it, which is why overclocking is feasible but inconsistent.

View on HN

This is also common a common practice with garments, bakery items, fruit, racehorses, and children.

Wait, children?

Should be an XKCD comic

Makes sense, no?

Some children are just smarter, more athletic, or better looking than their siblings.

I think they are referring to this part:

> It was the same chip, just that some came out the oven better than others.

Not the part about selling.

There are shops that buy CPUs in bulk and bin them themselves for suitability for overclocking and then sell the top performing ones at a premium.

I got my i9-9900k this way, they binned it for 5.1GHz and it had been working 24h non stop for a few years (it is still working fine, but I no longer have a need for it to run).

Silicon Lottery shut down two years back. It isn't clear to me whether manufacturer binning got tighter, overclocking just doesn’t work anymore, or auto-boosting are their lunch.

These days chips with different amounts of cores or cache may derive from the same mask. If one or more of the cores or banks of cache are defective, they're disabled.

Intel/AMD do not test every single die. Once you get stable output from the foundry you learn a max clock/wafer location distribution.

That's interesting. I always thought each one had a basic pass-fail test after packaging.

Reply on news.ycombinator.com