Intel users: prepare to lose 5-30% of performance

[spacein_vader]

TL;DR - Prepare to lose 5-30% of your CPU performance across the board, regardless of OS.

An embargoed bug in all modern (last decade,) Intel chips has shown a fundamental design flaw that allows unauthorised access to the kernel (the deepest, darkest and in theory most secure area of your computer,) that cannot be patched out of the microcode or drivers. Instead each x86 operating system (so Linux, BSD, Windows MacOS…) will have to patch in a workaround for it. Apparently MS has one ready to go for the upcoming patch Tuesday and the Linux kernel is getting updated with a speed and total lack of patch notes that is unprecedented as well as MS and Amazon warning customers of its cloud products of upcoming downtime for system updates.

These workarounds do work but they cause a speed penalty of 5-30% for all tasks which involve kernel access, which is basically everything. The workarounds are permanent, Intel cannot fix the underlying problem in existing chips.

AMD CPU’s are unaffected. If you want more technical details try this article https://www.theregister.co.uk/2018/0...u_design_flaw/

My view as someone with Intel chips is I’m mighty annoyed. If they lose the 5% speed they lose any advantage they had over Ryzen, if 30% is closer to the mark then that’s a hell of a downgrade. There’s no way they won’t be getting sued for this. Good day to be either AMD, an AMD user or a class action lawyer though. Even 5% difference in performance is a huge deal with massive costs for the likes of Azure and AWS.

[adidan]

What's the betting any workaround will impact W10 least of all so everyone running any other MS OS has to switch to W10?

[smargh]

My guess is that there will be a HKLM reg key, or maybe a boot flag to disable the workaround, as is currently planned for the Linux patch. If a system has already been compromised then the patch can just be uninstalled/modified anyway.

[Lanky123]

Looking forward to a data centre or ten emptying their Xeon collection onto eBay then. Every cloud...

[Fury559]

Which models are affected?

[ik9000]

Which models are affected?

ALL of them since core architecture onwards iirc. this broke a few weeks ago but it was unclear how readily it could be patched. if it is true that a firmware fix is not viable then that is a bigger problem than we were led to believe at the time.

[Xlucine]

AMD (especially the epyc team) have probably popped the champagne already. Is it just me or has there been a lot of serious intel bugs over the past year or so? I suspect no-one saw any similar issues in the construction cores because no-one bought enough of them to spot the issues...

[Saracen]

What's the betting any workaround will impact W10 least of all so everyone running any other MS OS has to switch to W10?

Well, not everyone is switching to W10, even if it is 30%.

What I might do is switch all non-essential systems to my air-gapped network, and just not patch. After all, a good chunk of my systems are already and most of those that have a net gateway don't actually need it, and certainly not at the price of going W10 .... which aint 'appening.

[Sam__]

Surely the performance hit can't be that bad. If so then Intel are completely done for.

[DanceswithUnix]

Surely the performance hit can't be that bad. If so then Intel are completely done for.

Yes it can, the switch from user to kernel needs to be as smooth as possible as it can happen tens of thousands of times per second.

As for Intel being done for, they have weathered stuff like this plenty of times. I expect by March you will hear people saying how they buy Intel for the quality again. Most likely outcome is that people will go out and buy new Intel processors, so Intel will make a huge profit from their mistake.

Edit: It will be interesting to see if Vulkan/DX12 hurts more or less than DX10/11/OpenGL from this. In Windows, the graphics drivers were put in the kernel back in NT 4.0 days. Either way, I can imagine games being impacted.

[adidan]

Well, not everyone is switching to W10, even if it is 30%.

TBH I was mulling over the exact same thought.

[Corky34]

Yes it can, the switch from user to kernel needs to be as smooth as possible as it can happen tens of thousands of times per second.

Worth pointing out the key word there is "can" as not everything is going to be effected, performance hits are going to be highly dependent on how much user to kernel switching is done, games for instance do very little to none, filesystem operations do lots, phoronix have done some early benchmarking on Linux if anyone's interested.

This bug sure puts Brian Krzanich dumping of stock options a couple of weeks ago in a new light.

[DanceswithUnix]

games for instance do very little to none

On Linux, yes. Note that it is traditional in Linux to run graphics drivers in userland only using the kernel for initial setup, but Windows has them in kernel space. Given that Linux gaming generally sucks by about 30% vs Windows performance, this could be quite interesting, depending on how much of the driver is done in user space libraries vs talking to hardware in the kernel.

Edit: Wonder if the price of Intel silicon will drop from this.

[Agrippa]

Can't say I've noticed any performance impact from this. Gaming is the only area where I'd likely notice any slowdown, but frames per second and general performance appears unchanged.

Storm in a teacup?

[DanceswithUnix]

Can't say I've noticed any performance impact from this. Gaming is the only area where I'd likely notice any slowdown, but frames per second and general performance appears unchanged.

Storm in a teacup?

The online servers used by games are apparently seeing huge pain. They will be file access and network stack heavy, so not a surprise really. If they are based on public cloud servers, then they absolutely have to be patched as well.

[smargh]

Can't say I've noticed any performance impact from this. Gaming is the only area where I'd likely notice any slowdown, but frames per second and general performance appears unchanged.

Storm in a teacup?

Have you done the BIOS update as well? And checked the registry flag is set? The BIOS update exposes some CPU things to fix one of the things & enable full mitigation.

(I use "things" because it's easy to to get confused about all the different aspects of Meltdown/Spectre and which things fix the bad things & also need other things doing)