Hacker News Reader

You seem to miss the huge list of speculation bugs with mitigations that take a variable chunk of performance away with no buyback or return program in place. VW diesel gate was less of a blunder than the Intel speculative execution saga and one cost a company $33 billion.

View on HN

Wait, really? My understanding was that, whereas VW intentionally cheated to fool regulators, Intel (and others) got blind-sided by a creative attack against a fundamental, well-established technique for gaining performance. These seem like very different problems.

Note: I haven't followed the "saga" closely so maybe there is more to it that I'm not aware of.

Speculative execution is fine, speculative execution across protection domains is not. So VW cheated for better milage, Intel cheated for better performance.

"Cheating" implies something sinister. Far more likely, they just missed it.

Modern CPUs are at the very edge of what humans can design and build. Hundreds of verification engineers try to make sure they work. Not one of them will have a whole picture of the entire machine in their head, or really understand all of its requirements. The very best, hardest working, most diligent and dedicated teams will still ship a CPU with bugs. Perfection is simply not possible in this domain.

Reply on news.ycombinator.com