Remix.run Logo
▲ SV_BubbleTime an hour ago

I do embedded. I’ve always marveled at how ludicrous that “real software” is. I have truths. I have registers that will never be or act differently, I can make pretty damn good assumptions that no one will put this code somewhere else’s etc.

In the past few years though, it’s been creeping into my domain too.

▲bumby an hour ago | parent [-]

>I have registers that will never be or act differently

Even this needs a qualifier. Radiation exposure can cause bit flips. Certain aerospace applications must address this, and even some earth-bound applications at high altitude.

▲LPisGood an hour ago | parent | next [-]

At high altitude or high deployment numbers. Hyper scaler data centers measure these “one in a million” anomalies in thousands of times per houe.

▲leftnode an hour ago | parent | prev | next [-]

Yeah, don't we have error correcting RAM for this reason, here on Earth?

▲Recurecur 5 minutes ago | parent [-]

Sure, but RAM isn’t the only place a bit can flip. Registers are a thing. IBM had error correction in Power architecture CPU registers, not sure where AMD, Intel, or Apple (lol) are on that issue. IIRC register bit flips first came up with Sun SPARC hardware, which Sun mitigated, but of course SPARC isn’t used too much these days…

A quick AI-assisted search informs me that neither AMD nor Intel document any on-CPU error correction, though perhaps there is some. Apple currently has zero systems that even use ECC, which seems insane to me given they’re shipping systems with up to 512 GB of RAM.

This seems to be an issue that’s falling through the cracks again. I guess nothing will be done until there’s a critical failure due to a bit flip, and verifying that was the cause for a given problem is difficult.

On ECC, Apple needs to Think Different (tm). Maybe it should introduce actual “Pro” machines…

▲marcosdumay an hour ago | parent | prev | next [-]

Besides, chip bugs are famous since the Pentium, and unreliable circuits are the normal way to describe circuits, you have to add notice when you get a reliable one.

▲skydhash an hour ago | parent | prev [-]

That's why we do requirement gathering and analysis. Engineering is always about tradeoffs and things that matters in certain context don't in others. The issue is that in a lot of orgs, while there's a record of decisions, we don't usually track assumptions and the overall system is full of contradictions. Also there's no overall keeper of knowledge (either one person or a team) that periodically ensure that contradictions are kept at minimum.