Ask Lemmy

So many people here don't really know what they're talking about. Chip companies can't cut corners on stability because of the amount of money that goes into everything around the chips is huge. If you think of a PC motherboard and you look at how many components are on that board now, imagine if one of the company's cheaped out on a chip and didn't bother testing it before sending it out and the only time they found out it was bad was when they finished making a board. The cost of finding, repairing and replacing those components far exceeds the pennies that they save by cheaping out.

The real answer is that their customers are a user of the chips and the cost of a bad part is massively more expensive than a tiny savings in manufacturing the chip.

The only place they cheap out on parts is in things that are standalone and dirt cheap such as an RFID chip. I've seen those get manufactured without test and only at the very end after they've been assembled to their antenna they then test and reject the ones that fail. They get away with it there only because the cost of the chip and antenna are so cheap.

Imagine Apple putting a sub-tier component in their phone and having to recall 10% of their phones because of it. It's unheard of. The people who buy the chips are usually companies that use them and they have a very low threshold for bad components. While it's true, some parts are binned for performance, they are never binned for quality.

Note - I've been involved in the test of semiconductors for the past 20+ years making load boards (package test) and probe cards (Wafer test) for many different IC manufacturers.

They used to. Or rather, the industry used to. Most of those players have since folded, largely because of them becoming known for crap quality.

Ummm they do? If you want shit chips that don't work you'll find loads of junk on eBay

They kinda do (look up binning - not exactly the same thing though, there is still QC there) but the problem with putting out unstable chips is that, unless all the other companies do the same thing, people will just start buying from the companies that make the chips/devices that function reliably. Nobody wants to use a computer that crashes all the time or is otherwise nonfunctional.

I would think even tiny decreases in stability would drastically increase the frequency of system crashes. Software is generally made expecting total consistency in computer instructions like, for example, integer arithmetic. Any interference with that would probably not be cost-effective on the part of the chip manufacturer.

Two reasons (personal lore. No proof) why they arent doing it as much:

1. Reputation (high failings = Low sales volume, higher returns, decreased reputation)
2. Higher vlume of returns. They arent cheap.
   And if they say "free return" they just charged you and every other order 20 cents more to compensate for the 10 returning packages of 100 shipments.
   Nothings free.

First of all, they do do this. The AMD "tri-core" chip was a quad core that failed QA. Then most slower chips are in fact from faster chips but also failed QA and have to be underclocked.

However, the real answer is that because if you get a bit flipped, while most things are recoverable, there's a good chunk at the hardware level that is unrecoverable. You've seen this with blue screens. Chips like that wouldn't sell. Bluescreens and panics aren't acceptable to people.

This was Intel's 13000 and 14000 i7/i9 lines

I remember when Intel made Pentium CPU's that had a small math error in some very specific floating point calculations. They were so afraid to damage their reputation (which was still excellent at the time) that they offered every Pentium owner across the globe (including me) a free new Pentium CPU without the bug, shipped to us at their expense, and even sending out a courier to pick up the old CPU (again for free) a few weeks later when we had time to swap them. That was basically the opposite of what you're suggesting.

Happens all the time... Back in the day if you had a chip with a failed math co-processor, you just sold it as a cheaper version without it. That kind of thing.

They kind of do, or at least used to.

If memory serves, they would take higher-end chips that didn't pass QA for that product line, disable some cores or whatever, and sell it as a lower-end chip.

You don't think they're doing that as I'm writing this??

Most of them are easily replaced. You won't notice a difference if it's an AMD, intel, seagate, kingston, etc that made your drive, ram, CPU.

Because they can be replaced very easily they can't enshittify it. The tech giants that enshittify has made themselves very hard to replace in one way or another.

Isn't that Intel's MO?

It's already common for chipmakers to disable cores entirely because of a single mistake. With 100 Billion+ transistors, even a doubling in error rate would 100fold the number of wasted dies.

Making the stencils is already expensive there's no real way to cheap out on it.

because it would tank their reputation and nobody would buy their chips.

you have to understand, chips are commodity items. they aren't 'sexy' or marketed heavily. people don't choose one chip brand over another based on how 'sexy' it is'. They mostly just don't care as long as it works. technology is at this point is pretty much a commodity/utility for most people and most use cases apart from high end, low volume applications like PC gaming or etc.

the reason like a car manufacturer can do this is because the 'image' of the car is so appealing to people they will overlook how shit it is and how poorly it's built. computer chips don't have sex or image appeal in this way, so they really can't afford to produce a shitty unreliable product.

consumer grade chips are already 'low quality' compared to the server end stuff anyway. they have cheaper tolerances and lack error correction and other features that are necessities for service-level computing. not to mention the really high end stuff you find in super computing and military/industrial hardware, etc.

Make games. Tabletop and video games.