Is It Worthwhile Running Intel Alder Lake With mitigations=off?

[perpetually high]

Thanks for the heads up, tildearrow.

[tildearrow]

You obviously have some hidden agenda being so aggressive when all I asked you was for sources. Take the best of care, avem. Mr. June 2021 join date.

Hey there.

To tell you, he's actually birdie. Mr. July 2008 join date.
He just thinks that hiding under another account will allow him to insult and troll better, but we all know avem is birdie.
It's best you seriously stop discussing with him, as he only wants conflict all the time.

[perpetually high]

You obviously have some hidden agenda being so aggressive when all I asked you was for sources. Take the best of care, avem. Mr. June 2021 join date.

[avem]

Since you wanna be disrespectful, let's go. I'm talking about microcode updates, AGESA changes, kernel changes, scheduler changes, literally everything that contributes to mitigating the performance impact of all the security mitigations in the kernel.

I highly, highly doubt you're even a developer or someone that tracks the changes to know what I'm even talking about, let alone worth a further discussion. Take the best of care.

Now on to your BS: we are talking about transient execution attacks and the associated performance loss. Have you actually ever tested the performance impact of AMD microcode updates? Zen 3 CPUs have seen just two (!) updates for their entire lifetime. It's not like Intel which releases them almost quarterly.

Kernel changes and scheduler changes? I presume you can provide lots of citations (patches/changes/etc) for the past 12 months, cause it must be trivial, right? Nothing? Just a verbal assault?

Literally everything fixes a 7% performance loss?

If you want to be a level above me, smart, unbiased, patient and cool, start providing fucking solid facts and raw benchmarks (not just four but a couple of dozens), 'cause otherwise you're no better than a buffoon. I've given you a test by Michael, publicly available and verifiable.

Speaking of your own test results.

OSBench
Test: create files

on 10.8
off 10.22

That's 5.6% performance loss. Right there. Not 0.5%.

To quote you:

"I highly, highly doubt you're even a developer or someone that tracks the changes to know what I'm even talking about, let alone worth a further discussion. Take the best of care."

[perpetually high]

Have Zen 3 CPUs magically become faster in the meanwhile? WTF are you talking about? Yeah, there've been some odd improvements in the kernel but nothing too serious.

Since you wanna be disrespectful, let's go. I'm talking about microcode updates, AGESA changes, kernel changes, scheduler changes, literally everything that contributes to mitigating the performance impact of all the security mitigations in the kernel.

I highly, highly doubt you're even a developer or someone that tracks the changes to know what I'm even talking about, let alone worth a further discussion. Take the best of care.

[avem]

December 2020 is far from recent. That’s exactly one year ago. Lots has changed. We’ll have to get a refresh of these benchmarks. Regardless, thanks for following up avem

Have Zen 3 CPUs magically become faster in the meanwhile? WTF are you talking about? Yeah, there've been some odd improvements in the kernel but nothing too serious.

[avem]

What evidence do you have for this? In this case, Alder Lake has a host of mitigations baked in, and we have no theoretical 'clean' silicon to test against to see how efficient the mitigations are.

Are you sane? Previous gen Intel CPUs show quite a performance loss of up to 90% in certain tasks, while for ADL it's below 1%. What further evidence do you need other than benchmarks?

[Alex/AT]

Given 'mitigations=off' even gives worse performance at places while it definitely should not by design, I start to suspect another cheat buried somewhere in the new hardware. Just wonder how many time will pass before it's discovered and exploited.

Looks like Intel has done a spectacular job of addressing performance regressions related to the use of mitigations

Yeah indeed, looks the same to me. Last time a similar spectacular 'addressing performance' job skipping privilege checks resulted in Meltdown L1 timing exploit relying on out-of-privilege access being discovered, so can't be anything but sceptical here about the new findings.

[NobodyXu]

With due respect to Michael, imo is more important to know if the mitigations are actually still needed, considering is a new architecture and very likely mitigations are already in place on the design.

Maybe it would be a good idea to disable these migrations by tinkering with the CPU microcode?

For the old CPU we can downgrade microcode, but for the new ones… perhaps it can be disabled in BIOS?

[bug77]

Thanks for working out that math on that, I was curious myself lol. And yeah I didn't mean to say "zero effect" on Zen 3 since I did see *a* difference but not as much as Alder Lake, which is really impressive by Intel.

The numbers thrown around in comments are deceptive. Mitigations' impact has always been highly workload dependent. Most scenarios unaffected, select few hit rather significantly. Averages tell squat in this context.