An instruction added as a curiosity / micro-optimization back in 2010 is now very important.

Previously: "PCID is now a critical performance/security feature on x86" https://news.ycombinator.com/item?id=16094349

Interesting that it's more substantial on older CPU's. The cynic in me wonders if that's a way to force an upgrade from CPU's that normally would be fine for most typical office use.

This has absolutely destroyed Intel's reputation and handed a massive PR win to their competitors (including AMD). Hard to imagine any normal company cutting out their own feet so willingly, this will cost Intel a lot more than it will make them in short term revenue.

I'm not sure how they could have come out of this cleanly other than waving a wand and not having it happen? They pretty clearly owned up.

They sat on it for over quarter and gave no indication that any sort of trouble was coming up. The CEO even sold stock and made a profit after being made aware of the issue, but before having the issue be announced publicly. I would not classify that as clearly owning up

Did we see the same PR announcement?

It was misleading in multiple ways.

Really? I thought they were still denying that this is a flaw, and not, “working as intended.”

Do what i do, when my android phone constantly flashes a light & sound notification after finnishing loading the battery- call your software representetive.

It really helps against insomnia, if you wake some poor android developer halft the world away in the middle of the night to suffer with you - for his implementation of planned obsolence.

Two souls united in burning hatred.

>>> If you're using a 4th-generation Core or older CPU, however, you'll see "more significant slowdowns,"

With server CPUs always being a few generations behind the desktop, this is gonna have massive impact on datacenters.

Heh. Have you seen the Phoronix benchmarks?

https://www.phoronix.com/scan.php?page=news_item&px=KPTI-Ret...

E3-1280 v5 goes from 32278.00 Apache requests per second to 25399.47 Requests per second. In other words, a Skylake Xeon performs like an Fx Bulldozer after the patch. And Skylake has the PCID instruction too.

Although to be fair, static-file hosting is about as I/O bound as it gets, and these patches affect file I/O the most.

Compile time slowdown looks pretty sad too. -21% is quite bad.

Is there any idea when this will be fixed in hardware?