Hardware bug in Skylake and Kaby Lake : Unsafe Hyper-Threading

[tom_w141]

23 minutes ago, Morgan MLGman said:

Only 2 pages of the thread? If it was about Ryzen and not Skylake/Kaby Lake it'd be at least 10 pages long

As for the thread itself, how does this affect users such as me running a 6700K under Windows 10? What are the risks and what triggers the bug?

Actually if you make a thread negative toward or Intel or implying they aren't perfect the blue knights storm in on horseback pretty fast so you actually get a lot of posts  of course the chance those posts have anything to do with the thread topic is slim but the posts will be there nonetheless  

 

On topic: This is very unlikely to affect the every day user as the exact condition to reproduce requires certain instructions in a specific order.

[SCHISCHKA]

1 minute ago, tom_w141 said:

Actually if you make a thread negative toward or Intel or implying they aren't perfect the blue knights storm in on horseback pretty fast so you actually get a lot of posts  of course the chance those posts have anything to do with the thread topic is slim but the posts will be there nonetheless  

 

1 minute ago, Morgan MLGman said:

Thank you for the explanation :3

watch as this post grows exponentially just because someone mentioned Ryzen

[tom_w141]

1 minute ago, SCHISCHKA said:

 

watch as this post grows exponentially just because someone mentioned Ryzen

Prepare to be told about Intel performance vs Ryzen

[tom_w141]

13 minutes ago, Morgan MLGman said:

I somehow never manage to catch those same people comment on issues with Intel CPUs or if they do, those issues are "suddenly" irrelevant while they may actually be more severe Hypocrisy at its finest.

^ While I agree this is not a big deal and if you have been fine you will probably carry on being fine, the above is so true.

[LAwLz]

9 hours ago, mariushm said:

Users should disable hyperthreading until Intel comes with a microcode update.

No they shouldn't, unless they are running some code which is known to cause the issue and/or are running very mission critical code.

So for 99.999% of users on this forum, they should not disable HT. Disabling HT would most likely cause more "damage" to users on this forum than leaving it on would.

 

9 hours ago, Alaradia said:

has intel replied to these allegations?

They have released an update for it already.

 

9 hours ago, ARikozuM said:

I wonder if this is a problem with gcc, unless I'm mistaking an issue with Ryzen with this one. 

Not GCC related.

 

9 hours ago, DocSwag said:

Um so what do I do if I'm on windows and have skylake? Is there nothing I can do?

Here are the things I would recommend (in order of recommendations):

1) Don't do anything unless you have ran into this issue (and don't be too quick to attribute issues to this bug).

2) Check for BIOS updates and see if it contains the fix.

3) Disable Hyperthreading.

 

9 hours ago, NumLock21 said:

There are other issue that no one has heard about it yet.

Yeah... I fear the day LinusTechTips find AMD's, Intel's and Nvidia's documented workarounds for hardware issues.

For those interested, here is volume 16 of Intel's workaround documentation for their GPUs. Yes, that's a 44 page long document listing issues and how to avoid them. I couldn't find any newer version, hence why it's from 2016. AMD and Nvidia have similar documents published to developers.

 

 

9 hours ago, Cool Guy said:

Maybe you just haven't used the processor enough to cause it to hyper thread to an unsafe level.  It's not like you're losing any gaming performance disabling hyper threading.  And if you're rendering and using hyper threading on a constant basis, your chance of encountering this bug is quite high I would imagine.  Your best bet would be to just disable it. 

 

It's also possible that the effects of this bug might cause a system freeze or crash, but not instant death of the processor.  Maybe just some permanent damage.  Do you want to risk your processor getting damaged (probably is already if you use hyper threading on a constant basis)?  I don't think your comment is very wise.  It's your decision.  But think it over carefully. 

Stop spreading FUD please.

Where did you get the idea that this will brick processors or cause some permanent damage from?

 

8 hours ago, imreloadin said:

So since this is a Linux bug does that mean Windows isn't affected?

Seems like it's a hardware issue, so other OSes should be affected too.

 

8 hours ago, AresKrieger said:

holds little water due to that fact, additonally this doesn't look like a source we can verify so I'm dismissing it until proof emerges

8 hours ago, Sierra Fox said:

unless Intel acknowledges this then I wouldn't worry. unverified claims. and given how long skylake has been out and this is now just an issue, seems a bit off.

The mail is from a Debian developer, and Intel has released a patch for it (and many other issues), so it is legit. It's good to be skeptical, but dismissing something like that is not a good idea.

 

 

8 hours ago, DocSwag said:

It's possible that it's a universal issue, but something within Debian is more likely to trigger it.

 

This is something that is fixed with a microcode update not a software update so definitely universal, some oses just might be a lot more likely to trigger it.

8 hours ago, ARikozuM said:

I'm going to have to assume so given that both Ryzen and Skylake/Kaby Lake show the symptoms.

8 hours ago, AresKrieger said:

That is also possible, frankly I'm not worried about this issue and wouldn't be even with a skylake cpu it seems like an isolated problem additionally until someone verifies it I'll treat it as false until proven true (since information is generally false online especially info without solid sources)

It's not really like Debian users are more at risk either. Like the mail says, it was sent out because some Debian stable users had run into this issue. The issue was discovered by some OCaml users. OCaml is specifically designed to be very stable and used for a lot of things highly sensitive to downtime, for example Bloombergs risk analysis programs.

So this info was not really meant for LTT-type users. It basically went like this:

User who manages millions or billions dollars worth of data: "Oh, there is a risk that something might go wrong with my program."

Debian mailing list: "Hey everyone, this is a potential issue. Here is how to fix it."

This thread: "I play Minecraft on my computer and this is very serious to me."

 

 

8 hours ago, NvidiaIntelAMDLoveTriangle said:

Oh Debian, just because your OS is broken, doesn't mean the hardware is too.

This is not a Debian issue.

 

 

1 hour ago, Morgan MLGman said:

Only 2 pages of the thread? If it was about Ryzen and not Skylake/Kaby Lake it'd be at least 10 pages long

Yeah, it's really weird that issues that has affected probably a handful of people and appears to have already been fixed don't get as much attention as issues that affect millions of users. It gotta be because of fanboys, right? Clearly they should both get as much attention, because what matters isn't what the issues are or how wide-spread they are. What matters is which brand messed up.

/sarcasm

 

[Morgan MLGman]

1 minute ago, LAwLz said:

Yeah, it's really weird that issues that has affected probably a handful of people and appears to have already been fixed don't get as much attention as issues that affect millions of users. It gotta be because of fanboys, right? Clearly they should both get as much attention, because what matters isn't what the issues are or how wide-spread they are. What matters is which brand messed up.

/sarcasm

You know I was talking about priciples and hipocrisy in people fanboying one or the other company, not about the issue itself ;_;

 

/misunderstanding

[AresKrieger]

6 minutes ago, LAwLz said:

Debian mailing list: "Hey everyone, this is a potential issue. Here is how to fix it."

This thread: "I play Minecraft on my computer and this is very serious to me."

 

Well yeah as I said it seems to be an isolated issue and even if it wasn't given that newer bioses don't have the issue it is rather irrelevant, the reason people seem to be bothered by this is that the OP made it seem like a problem and the post itself is too long for most people to read/comprehend fully

[Misanthrope]

8 hours ago, ivan134 said:

Average framerate isn't everything. Getting 300 fps (redundant, but that's not the point) in some areas and dipping down the 110 fps (this makes hitscan heroes hard to play) in others in Overwatch, with all settings on low, is the reason I'm switching to an i7. Hyperthreaded CPUs give better minimums in CPU intensive games. Then there is the issue of stuttering. Fallout 4 and Battlefield 1 are not fun to play with an i5. Fallout 4 made @App4that buy a 4790k.

 

  Reveal hidden contents

 

 

Minimum framerate also isn't that much worst without hyperthreading. I get what you're saying but you're maybe 1 intensive setting away on a handful of games from spending 100 bucks. For most gamers it's far better to put that towards a better GPU than the hyperthreading.

 

But if budget isn't an issue of course: Get a 7700k, put it on water and have fun.

[porina]

1 hour ago, Morgan MLGman said:

I only worry because I still don't entirely know what that bug means for someone like me running an OC'd 6700K under Windows 10.

Think about how many i3 and i7 Syklake CPUs have been sold in about the 2 years they've been out. If this was a significant problem you can be sure we would have heard about it earlier. It seems that a highly specific condition is required, and if you haven't already encountered it in something you use, it would remain a low risk. Software might change, or worse, someone might add it to malware, but right now it isn't a big deal. Just keep doing what you're doing. If you suspect something happening might be this bug, only then disable HT and see if that resolves it. Then turn it back on and see if it comes back. Obviously there are many other things that could be happening, such as HT on makes more demands of the CPU, so OC stability may be more challenging for a given running condition than without.

 

As publicity is building up around it, it is more likely a fix will be pushed out one way or another in the near future.

[Dabombinable]

4 minutes ago, Misanthrope said:

Minimum framerate also isn't that much worst without hyperthreading. I get what you're saying but you're maybe 1 intensive setting away on a handful of games from spending 100 bucks. For most gamers it's far better to put that towards a better GPU than the hyperthreading.

 

But if budget isn't an issue of course: Get a 7700k, put it on water and have fun.

Or if budget is an issue, a locked i7 which is better for games than any i5....

[Morgan MLGman]

1 minute ago, porina said:

Think about how many i3 and i7 Syklake CPUs have been sold in about the 2 years they've been out. If this was a significant problem you can be sure we would have heard about it earlier. It seems that a highly specific condition is required, and if you haven't already encountered it in something you use, it would remain a low risk. Software might change, or worse, someone might add it to malware, but right now it isn't a big deal. Just keep doing what you're doing. If you suspect something happening might be this bug, only then disable HT and see if that resolves it. Then turn it back on and see if it comes back. Obviously there are many other things that could be happening, such as HT on makes more demands of the CPU, so OC stability may be more challenging for a given running condition than without.

 

As publicity is building up around it, it is more likely a fix will be pushed out one way or another in the near future.

Yeah, I know it shouldn't be an issue. I just got a bit annoyed when I've read the thread at first, Skylake already had a quite infamous freeze bug some time ago and who knows how many else 'hidden' ones are there, I didn't pay this much for my 6700K to have issues with it ;-; (Though I didn't have any yet, luckily)

 

I'm eagerly waiting for Cannonlake to truly start the competition between Intel and AMD. As X299 isn't really a successful platform launch (rushed, overpriced and limited in features that require to pay even more to 'unlock' some that are already there, ohh, and paying at least 1000 USD to get more than 28 PCI lanes) it's thanks to the 6C/12T mainstream i7s and 6C/6T mainstream i5s that the CPU competition will really heat up.

 

 

Though seriously, f*ck Intel for delaying progress in the CPU segment. What's funny I really hope they do good despite that so the rivalry gets more and more exciting and the end-users (us, gamers & enthusiasts) benefit from it even more

 

8 minutes ago, Dabombinable said:

Or if budget is an issue, a locked i7 which is better for games than any i5....

Fully agree. 4C/4T is just not good enough for high-end gaming and supporting high-end GPUs anymore

[Misanthrope]

13 minutes ago, Dabombinable said:

Or if budget is an issue, a locked i7 which is better for games than any i5....

A locked i7 also has lower frequency than an unlocked (so it's not just the overclocking) one so actually not really: For the price even without overclocking an unlocked 7600k will be noticeable better once overclocked than the 7700 nonk. Yes the non k part will have the edge on minimums but as shown, it's slightly better while a 7600k at 5ghz shows a more significant gain in IPC than just adding hyperthreading.

 

This is just going intel vs intel btw: the 7700 makes just about 0 sense since you can get an R5 1600 and a motherboard for around the same money AND overclock those and match the 7700 in games and surpass it by a lot in productivity so no, I wouldn't recommend people get a locked i7 at all at this point that should be only for people not wanting to upgrade their entire platform (i.e. someone on an i3 that doesn't wants to upgrade mobos) 

 

 

[Dabombinable]

18 minutes ago, Misanthrope said:

A locked i7 also has lower frequency than an unlocked (so it's not just the overclocking) one so actually not really: For the price even without overclocking an unlocked 7600k will be noticeable better once overclocked than the 7700 nonk. Yes the non k part will have the edge on minimums but as shown, it's slightly better while a 7600k at 5ghz shows a more significant gain in IPC than just adding hyperthreading.

 

This is just going intel vs intel btw: the 7700 makes just about 0 sense since you can get an R5 1600 and a motherboard for around the same money AND overclock those and match the 7700 in games and surpass it by a lot in productivity so no, I wouldn't recommend people get a locked i7 at all at this point that should be only for people not wanting to upgrade their entire platform (i.e. someone on an i3 that doesn't wants to upgrade mobos) 

 

 

I went from an i5 4440 (locked at 3.3GHz) to this 4790K. Even when I ran it underclocked to 3.4GHz (I was using shitty TIM+the stock cooler from the i5) it was far better in games than the i5 and that's with DDR3 1333. I know the difference in performance when it comes to games between an i5 and i7-even locked i7 are better than i5.

[porina]

28 minutes ago, Morgan MLGman said:

Yeah, I know it shouldn't be an issue. I just got a bit annoyed when I've read the thread at first, Skylake already had a quite infamous freeze bug some time ago and who knows how many else 'hidden' ones are there, I didn't pay this much for my 6700K to have issues with it ;-; (Though I didn't have any yet, luckily)

The AVX bug was publicised and I think there were a lot of people blaming it for any problem they had at all. I doubt if the vast majority of them could really be traced back to that, more likely it was over optimistic overclocking or more subtly compatibility problems with very high speed ram. The conditions to trigger that were also very specific, in that it was detected running an old version of Prime95 with specific newer functions disabled and a certain job size. I run tasks similar to recent Prime95 24/7 in colder months, and had no problem that couldn't be best explained by OC and cooling.

[Misanthrope]

20 minutes ago, Dabombinable said:

I went from an i5 4440 (locked at 3.3GHz) to this 4790K. Even when I ran it underclocked to 3.4GHz (I was using shitty TIM+the stock cooler from the i5) it was far better in games than the i5 and that's with DDR3 1333. I know the difference in performance when it comes to games between an i5 and i7-even locked i7 are better than i5.

I seriously doubt that: probably a placebo effect on your end or some other issue. I have never seen drastically worst i5 results on any serious review.

[ivan134]

39 minutes ago, Misanthrope said:

I seriously doubt that: probably a placebo effect on your end or some other issue. I have never seen drastically worst i5 results on any serious review.

Wow. Even with so many 1st hand accounts and the benchmarks I showed, you're calling it placebo effect.  You said earlier that you are usually one setting away from not needing a better CPU. I told I'm on all low in overwatch. It seems your confusing GPU bound scenarios for CPU bound ones or do you mind telling me what setting should I turn down to stop the CPU from doing calculations hmm?

[Sniperfox47]

Also keep in mind, that CPUs can recieve microcode from the OS after boot... which is why one of the suggested options for solving it is to just install the microcode package from non-free. Why isn't this suggested very loudly? Because this is Debian and, like Fedora, they hate non-free.

 

I don't have time anymore to run through all of the many many many KBs on the Windows side, but I'd be really surprised in Windows 7/8/10 haven't had the microcode patch come through after boot.

 

So unless you're running very specific workloads in an offline build of Windows, any build of Fedora/Debian, or off of some UEFI kernel or Boot CD I highly doubt you'll ever encounter this problem.

 

https://wiki.debian.org/Microcode#Updating_CPU_microcode_within_Debian_.28Intel_or_AMD.29 for more info on how the Linux Kernel patches CPU Microcode. Windows has a similar mechanism.

[NumLock21]

3 hours ago, LAwLz said:

 

 

Yeah... I fear the day LinusTechTips find AMD's, Intel's and Nvidia's documented workarounds for hardware issues.

For those interested, here is volume 16 of Intel's workaround documentation for their GPUs. Yes, that's a 44 page long document listing issues and how to avoid them. I couldn't find any newer version, hence why it's from 2016. AMD and Nvidia have similar documents published to developers.

 

 

Well those are some. Did you see their updated specs sheet with x299 kabylake x and a list of erratas for 1151 and 2066.

 

Some that might be important 

Note that this does not apply to all class of cpus and some of the workaround is a bios or driver update.

• Complex Interactions With Internal Graphics May 
  Impact Processor Responsiveness
•  Processor Graphics IOMMU Unit May Report 
  Spurious Faults
• Processor May Run Intel® AVX Code Much Slower 
  Than Expected
•  Integrated Audio Codec May Not be Detected
• Executing a 256 Bit AVX Instruction May Cause 
  Unpredictable Behavior
• System May Hang During Display Power Cycles
•  Display Flickering May be Observed with Specific 
  eDP Panels
• Processor May Exceed VCCCore ICCMAX During 
  Multi-core Turbo
• Some Memory Performance Monitoring Events May 
  Produce Incorrect Results When Filtering on Either 
  OS or USR Modes
• Unpredictable System Behavior May Occur 
  When System Agent Enhanced Intel® Speedstep® 
  is Enabled
•  Display Slowness May be Observed Under Certain 
  Display Commands Scenario
• Processor Graphics May Render Incorrectly or May 
  Hang Following Warm Reset With Package C8 
  Disabled
• Using Different Vendors For 2400 MHz DDR4 
  UDIMMs May Cause Correctable Errors or a System 
  Hang
• Two DIMMs Per Channel 2133MHz DDR4 SODIMM 
  Daisy-Chain Systems With Different Vendors May 
  Hang
•  Camera Device Does Not Issue an MSI When INTx 
  is Enabled
•  Processor May Hang Under Complex Scenarios

 

[Misanthrope]

1 hour ago, ivan134 said:

Wow. Even with so many 1st hand accounts and the benchmarks I showed, you're calling it placebo effect.  You said earlier that you are usually one setting away from not needing a better CPU. I told I'm on all low in overwatch. It seems your confusing GPU bound scenarios for CPU bound ones or do you mind telling me what setting should I turn down to stop the CPU from doing calculations hmm?

Yes: your testing is worthless to me, still anecdotal evidence. Every testing I've seen supports what I've said: minor improvements on minimum framerates (more significant ones at the top end high refresh rate) not worth the extra money specially for most gamers at 60Hz monitors and such. 

 

The problem is that you refuse to acknowledge any nuanced position and want to go "Nope, bigger number means I'm right, other factors don't matter" 

[Morgan MLGman]

15 minutes ago, Misanthrope said:

Yes: your testing is worthless to me, still anecdotal evidence. Every testing I've seen supports what I've said: minor improvements on minimum framerates (more significant ones at the top end high refresh rate) not worth the extra money specially for most gamers at 60Hz monitors and such. 

 

The problem is that you refuse to acknowledge any nuanced position and want to go "Nope, bigger number means I'm right, other factors don't matter" 

Well, I for instance have seen a few Digital Foundry vids when the same gen i7 and i5 were compared, i5 loses up to 35FPS in games compared to an i7 (that's with a bottleneck scenario - testing at 1080p with Titan X Maxwell as a GPU) and the frametimes jump literally all around the place on i5s.

 

This most likely isn't as severe at low refresh rates as you mentioned, but the locked i7 is still a better buy in my book. Simply because the cost of such platform is lower as you don't need aftermarket cooling and a Z270/Z170 chipset mobo which both add to the cost of the platform and i5s are just slower in multithreaded workloads, what's more they are more likely to bottleneck GPUs even of 1070/980Ti level of performance ;_; i5s will just not last as long and will require an upgrade sooner. And you won't be able to multitask as well as on an i7, that's also worth noting as not everyone just games on their PC and nothing does more.

 

With that being said, a Ryzen R5 1600 wrecks both of those in terms of value (I think you already mentioned it) and basically in terms of performance too, as the trade of losing around 22% in singlethreaded tasks and gaining around 75% more performance in multithreaded ones is well-worth it in my books, especially that the singlethreaded performance is somewhere around Broadwell. Of course there are scenarios in which that singlethreaded performance will come in handy, but for an average consumer it's just not worth it.

[ivan134]

21 minutes ago, Misanthrope said:

Yes: your testing is worthless to me, still anecdotal evidence. Every testing I've seen supports what I've said: minor improvements on minimum framerates (more significant ones at the top end high refresh rate) not worth the extra money specially for most gamers at 60Hz monitors and such. 

 

The problem is that you refuse to acknowledge any nuanced position and want to go "Nope, bigger number means I'm right, other factors don't matter" 

Yes because I work for techspot. Lol.

[ARikozuM]

18 minutes ago, Morgan MLGman said:

Well, I for instance have seen a few Digital Foundry vids when the same gen i7 and i5 were compared, i5 loses up to 35FPS in games compared to an i7 (that's with a bottleneck scenario - testing at 1080p with Titan X Maxwell as a GPU) and the frametimes jump literally all around the place on i5s.

You should never test a CPU with a single GPU. To do so skews results as variance comes into effect on a program to program basis. 

[Morgan MLGman]

4 minutes ago, ARikozuM said:

You should never test a CPU with a single GPU. To do so skews results as variance comes into effect on a program to program basis. 

Well, that's probably the reason why they used the top-end (at the time) prosumer GPU available. Hard to benchmark with two GPUs as some benchmarking tools and games don't support multi-gpu setups, unless you meant two separate, different GPUs.

[leadeater]

50 minutes ago, NumLock21 said:

Processor May Hang Under Complex Scenarios

LOL what?

[MageTank]

That sad feeling when @LAwLz quotes nearly every post in the thread, except mine

 

40 minutes ago, Misanthrope said:

Yes: your testing is worthless to me, still anecdotal evidence. Every testing I've seen supports what I've said: minor improvements on minimum framerates (more significant ones at the top end high refresh rate) not worth the extra money specially for most gamers at 60Hz monitors and such. 

 

The problem is that you refuse to acknowledge any nuanced position and want to go "Nope, bigger number means I'm right, other factors don't matter" 

 

13 minutes ago, ivan134 said:

Yes because I work for techspot. Lol.

Mind if I jump in? I can totally fix this. Ahem. Hyperthreading helps*

 

*In certain titles

 

There we go. Problem solved. Oh wait, you wanted specific titles too?!?! Fine. Watch Dogs (both 1 and 2), Metro Last Light/2033, Dragon Age Inquisition, Crysis 3 (everyone saw this coming). Now, Battlefield 1 is an interesting test case, because when I tested it, I sometimes saw better performance with it on, and better performance with it off. I tried to remove all variables (testing on single player) but I could not for the life of me keep it consistent. That being said, every other title I listed has been a very clear, dramatic difference 8-10% (20% in the case of Watch Dogs 1 for some reason).

 

Now, these tests were performed back when I had a 4.2ghz 6700k. If need be, I'll gladly re-test on this 5.2ghz 7700k. Oh, and this new IMC lets me push my 2x16GB kit up to 4000 C18 now, so I can probably go back and test memory overclocking gains while I am at it.