Monday, June 26th 2017

Critical Flaw in HyperThreading Discovered in "Skylake" and "Kaby Lake" CPUs

A critical flaw was discovered in the way Intel implemented its simultaneous multi-threading technology, HyperThreading, on "Skylake" and "Kaby Lake" processors. Being a micro-architecture specific flaw, this could affect all implementations, from low-power mobile chips, to mainstream desktop, high-end desktop, and perhaps even enterprise-segment Xeon processors. At this time, there are no security implications of this flaw.

Intel chronicled this flaw in its micro-architecture errata "SKZ7/SKW144/SKL150/SKX150/SKZ7/KBL095/KBW095," and described it as follows: "Under complex micro-architectural conditions, short loops of less than 64 instructions that use AH, BH, CH or DH registers as well as their corresponding wider register (e.g. RAX, EAX or AX for AH) may cause unpredictable system behavior. This can only happen when both logical processors on the same physical processor are active." As an implication, Intel goes on to note that Due to this erratum, the system may experience unpredictable system behavior."
The HyperThreading flaw can be fixed through a micro-code update distributed as a UEFI firmware update. Typically, it becomes the responsibility of DIY PC motherboard, pre-built desktop, and notebook manufacturers, to distribute the update. The issue first came to light in a Debian Linux user mailing-list, although it affects all PC operating systems, not just Linux. Support groups of Debian recommend disabling HyperThreading in the UEFI setup programs of your computers as a temporary workaround, till the micro-code patch is applied. Disabling HyperThreading will reduce performance in multi-threaded apps. Source: Debian Mailing List
Add your own comment

98 Comments on Critical Flaw in HyperThreading Discovered in "Skylake" and "Kaby Lake" CPUs

#26
RejZoR
oxidized
And you're happy, are you not? :):):)
In a way, yes, yes I am.
Posted on Reply
#27
trparky
When I build my own machine I know every component that's in that build. I know the model numbers, I know the brands, and I know that it was put together the right way because I myself put it together.

As for OEM machines, I've looked at many OEM systems and most if not all of them use the cheapest stuff you can find. Cheap RAM, cheap PSU, etc. The kinds of stuff that I wouldn't be caught dead buying and putting into my system.
Posted on Reply
#28
Filiprino
I am not talking about AMD. I am talking about Intel.
What AMD does is irrelevant. Claiming more transparency from Intel than AMD is bullshit too.

Intel has not contacted with the Ocaml compiler developers who found that the problems they were having were rooted to Intel processors. That is shoddy behaviour.

As for processors with faults, yes, they all have bugs. Processor design, as of today, has much in common with software design.

Solaris17
lol?

Intel is more transparent with errata than AMD ever has. Its been there motto since forever! You can download the PDF errata documents of any intel CPU product family right from there website?

Silly consumers. This happens all the time. it just doesnt normally affect home users because the software is either IE resilient or it doesnt trigger the issue. Notice how the people that found it were running server teams with large data sets?

Does the forum consensus on TPU actually think CPUs are made without fault other than missing or wanted feature sets by consumers? every single CPU revision has a published errata.

This isnt shady at all. They do it with everything, if you can get your hands on/if the company publishes the documents. from chipsets to hard drive micro-controllers.

EDIT:: Lets take a look

http://support.amd.com/TechDocs/51810_16h_00h-0Fh_Rev_Guide.pdf

Looks like AMD does it too, and no plan to fix at that. Where is your god now AMD fans?

Here ya go. I took about 35seconds to shatter your delusions of purity.

http://support.amd.com/en-us/search/tech-docs#k=errata
Posted on Reply
#29
Shihabyooo
trparky
When I build my own machine I know every component that's in that build. I know the model numbers, I know the brands, and I know that it was put together the right way because I myself put it together.

As for OEM machines, I've looked at many OEM systems and most if not all of them use the cheapest stuff you can find. Cheap RAM, cheap PSU, etc. The kinds of stuff that I would be caught dead buying and putting into my system.
And they are -usually- obligated to fix it should anything happen to the machine due the quality of those components. So even they don't pass Corsair or Silverstone's marketability criteria, they do work, else the OEM has to fix them.

OEMs don't sell brand RAMs and cards with aftermarket water loops, they sell systems that work without the consumer bothering to assemble or maintain them. That's a completely different deal that what you get with DIY systems.
Posted on Reply
#31
RejZoR
Specific FMA issue on only very latest and newest CPU generation based on absolutely new architecture versus HT on TWO generations that have been on the market for years now. Granted, the scenario is a bit more specific than just using HT, but still. Sometimes it's better not to know what kind of bugged chips we use, otherwise you'd want to resort back to abacus...
Posted on Reply
#32
BrainCruser
RejZoR
Specific FMA issue on only very latest and newest CPU generation based on absolutely new architecture versus HT on TWO generations that have been on the market for years now. Granted, the scenario is a bit more specific than just using HT, but still. Sometimes it's better not to know what kind of bugged chips we use, otherwise you'd want to resort back to abacus...
I am more amazed by how well CPUs actually work considering how complex they are.
Posted on Reply
#33
Mescalamba
CAPSLOCKSTUCK
No problems here


Long live socket 1366..........:peace:
Yea Im still there too.

Only upgrade Im thinking about is liquid cooling. :D
Posted on Reply
#34
oxidized
RejZoR
In a way, yes, yes I am.
Yeah i think everyone knows fanboys in here by now :):)
Posted on Reply
#35
Mistral
Mescalamba
Yea Im still there too.

Only upgrade Im thinking about is liquid cooling. :D
920 for life!

Seriously though, I'm jumping on the Ryzen train as soon as the end of the year... or next year...
Posted on Reply
#36
Solaris17
Dainty Moderator
Filiprino
Intel has not contacted with the Ocaml compiler developers who found that the problems they were having were rooted to Intel processors. That is shoddy behaviour.
Why? Because an error that affects less than 1% of people using the CPUs only just now made the headlines? Isn't it only in the headlines because people like you make it big?

What about the 13 or more bugs in the AMD errata PDF I posted? I can go back in the TPU news articles and I can't seem to find any of the corresponding pitchfork threads.

How do you know? I mean you can clearly see the errata document for intel WITH the corresponding code that identifies that particular bug which was amended in via date stamp just this month.
Posted on Reply
#37
Vario
Would be nice to live in a world without rabid brand obsessed fanboys.
Posted on Reply
#38
oxidized
Vario
Would be nice to live in a world without rabid brand obsessed fanboys.
Yeah it would be.
Posted on Reply
#39
Prima.Vera
This Intel announcement couldn't be more vague and ambiguous even if they wanted to. What tf does "unpredictable system behavior" means??
Posted on Reply
#40
GreiverBlade
oh... well i should be glad that i have a i5-6600K then ...

not that HT is technically useful... for gaming ... (sometime it's even an hindrance :p )
Posted on Reply
#41
Frick
Fishfaced Nincompoop
trparky
When I build my own machine I know every component that's in that build. I know the model numbers, I know the brands, and I know that it was put together the right way because I myself put it together.

As for OEM machines, I've looked at many OEM systems and most if not all of them use the cheapest stuff you can find. Cheap RAM, cheap PSU, etc. The kinds of stuff that I wouldn't be caught dead buying and putting into my system.
To add to the answer above, they know what they are doing. There was a period when the capacitor plague indeed ran rampant on both motherboards and PSUs, but on the whole OEM machines just work. You have bad models obviosly, but the times of paperweight PSUs are gone now, largely. Dell has even used Delta in the past, at least for some models. FSP is a huge player here and they definitely know how to make long-lasting PSUs. And cheap RAM... Well I do that too! My computer can't run any 4×4 stickz in 1600Mhz, so I just have filthy cheap Ballistix sticks in them and none of my programs know it, and none of my programs care. The only metric you have is speed, and for 95% of the users that is completelt irrelevant as long as they have the amount they need. The same goes for the motherboard. Does it have the features they require? Anything else, any bling at all, is a waste of money for, again, 95% of the users. These are the people these machines are meant for.

And did you know no OEM board maker would even think about using RGB lightning?

And "properly put together" is an interesting term here. Is there such a thing as an "improperly put together" computer? No. Well, cooling can be improperly mounted and cables can have poor contact, but on the whole a computer is like IKEA furniture, just way, way easier to assemble. We don't build our own circuit boards. Most computer "enthusiasts" can't even solder; where's the damned heroism in that? Look I got eight different components and I managed to put them together in a system which does not allow things to be connected wrong unless I use saws and hammers! I am Hero, I AM SUPERIOR. I know what my graphics card is called; KOWTOW. I run memory at a specified speed; KNEEL. I have a motherboard with heatsinks that looks like things and it is black and red and surrounded by light and I can run my circuits faster than they are sold to me as; I AM THY GOD. My case is a tower of black perfection, designed by someone and made in China; I AM THE OFFSPRING OF HE-MAN AND WONDER WOMAN, I AM MASTER OR THIS UNIVERSE AND THE NEXT.

This turned out to be a bit of a rant I think.
Posted on Reply
#42
mcraygsx
First time Intel shocked me with their statement was when they told Consumers not Overclock "K" series CPU which was just recently and now Disable HT which does lower performance for some apps. That was the whole point of 7700K so we can get Unlock CPU with Hyper Threading. This almost seems like 7600K would've been a better deal. Hopefully micro code update is underway for those who are affected by this errata.
Posted on Reply
#43
Aquinus
Resident Wat-man
CAPSLOCKSTUCK
No problems here


Long live socket 1366..........:peace:
Meanwhile my 3820 keeps humming along. :cool:
Posted on Reply
#44
FordGT90Concept
"I go fast!1!11!1!"
Errata, as usual, are really fringe cases. There's no need to panic about it or overreact (disable Hyper Threading, unless you're actually seeing this problem). Just wait for new firmware (MSI hasn't released an update yet for my board but they did for unrelated issues a month ago), install it, and...

Wouldn't be surprised if most don't release updates because it's not a security issue.
Posted on Reply
#46
TheGuruStud
BrainCruser
No they aren't. They are keeping their head down, since Ryzen has a similar problem when using FMA instructions. GCC found them crashing the compiler and producing inaccurate code(deadly flaw).
Wasn't that already fixed weeks ago in a microcode update? IIRC, it needed just needed a tiny bit more power under that load.
Posted on Reply
#47
RealNeil
ssdpro
Do we have any cases of anything negative happening related to this? Skylake has been around a long time now and forums show remarkable stability with CPU related tasks.
Yeah, my 6700K has always seemed to be fairly bulletproof. If they issue a fix, I'll install it, but I'm in no rush for it.
Posted on Reply
#48
mtcn77
Solaris17
Why? Because an error that affects less than 1% of people using the CPUs only just now made the headlines? Isn't it only in the headlines because people like you make it big?

What about the 13 or more bugs in the AMD errata PDF I posted? I can go back in the TPU news articles and I can't seem to find any of the corresponding pitchfork threads.

How do you know? I mean you can clearly see the errata document for intel WITH the corresponding code that identifies that particular bug which was amended in via date stamp just this month.
'No fix' means fixed? I'm new to this genre...
Posted on Reply
#49
trparky
You know, of all people I would think people who hang out here would be the kind of person wouldn't be caught dead using a Dell or HP computer in their homes. Self-built or nothing at all. Building one's own computer was seen as a rite of passage, when one acquires their first merit badge of geekdom.
Posted on Reply
Add your own comment