Intel to address bug in Skylake Cpu's

[martin-w 3,235]

I wasn't contending that Prime95 is not using "new instruction sets" ... but, Prime95 operates valid instructions, a CPU should be able to handle those instructions.

That may be a bit simplistic. We are dealing with a new CPU architecture, and testing it with software not validated for that architecture. Neither of us are expert enough to state anything definitively. It may be that Prime, in this scenario, would not operate some of the newer instruction sets at all and operate other new instruction sets inappropriately. Or another scenario that neither of us are knowledgeable enough to hypothesize. Asus obviously believe something like that to be the case. And given that they are far more knowledgeable than me, and have carried out test that I haven't, I give their opinion consideration.

 

 

The above statement is not accurate and is dubious.

 

 

Why? Evidence of that? I'd be very surprised if Asus made a complete balls up with that statement, given their experience with the platform. Don't forget Rob, Asus get the CPU's quite a while before the consumer, and test a multitude of examples. They become experts, before we even sniff the CPU. Doubt they'd claim such a thing unless they had evidence it was the case.

 

I only quickly glanced at the article, but perhaps they were considering off-set or adaptive voltage, well known to be a concern when running synthetic stress tests?

[Guest]

 

 

Neither of us are expert enough to state anything definitively.

 

I not pretending to be an expert, but I can definitely say that is NOT the case ... we'll just have to disagree ... that's just not how CPUs are designed.

 

 

 

I'd be very surprised if Asus made a complete balls up with that statement, given their experience with the platform.

 

I wouldn't ... they're using scare tactics to get people to use RealBench instead of Prime95 ... CPU's have Thermal  protection, they shutdown before any damage (unless you disable it - which in some CPUs I'm not sure you can).  Asus is just trying to control the benchmark -- fair enough ... Asus, AMD, nVidia, Gigabyte, MSI, etc. have all at some point in time come up with "tricks" to make their products appear better than they really are ... no different than any Ad you see on TV (especially food ads) ... just marketing.

 

Anyway, I'm not suggesting everyone rush out and dump their Skylake CPUs, but I will suggest that once Intel release a "fix" run the same stress/performance tests as one did before the FIX just to make sure the numbers are the same and all is good.

 

Cheers, Rob.

[fakeflyer737 992]

How will the release a fix...in the forum of a BOIS or driver or what?

[martin-w 3,235]

I not pretending to be an expert, but I can definitely say that is NOT the case ... we'll just have to disagree ... that's just not how CPUs are designed.

Then I would have liked you to tell me how CPU's are designed and precisely why Asus are wrong. But I guess at this point, as you say, we'll just have to agree to disagree. :smile:

 

I wouldn't ... they're using scare tactics to get people to use RealBench instead of Prime95 ...

 

 

You finally said it!

 

Asus were saying this long before Real Bench was developed though. And Real Bench is free, Asus make nothing from it. Doubt they'd slag off the fine work of George Woltman just to push a utility they make nothing from. Just as some kind of marketing ploy. As you say, neither of us are experts, but there are a multitude of experts out there that would have cottoned on to the lack of validity in Asus's words.

 

CPU's have Thermal protection, they shutdown before any damage (unless you disable it - which in some CPUs I'm not sure you can).

 

 

 

Not exactly. Running any CPU just short of the point where the CPU throttles back, just short of TJ.Max, causes accelerated degradation. Not forgetting that many OC enthusiasts run Prime 95 for many, many hours. Same applies to the link you posted. Strictly speaking Asus are correct. Degradation is accelerated under those circumstances. 

[martin-w 3,235]

How will the release a fix...in the forum of a BOIS or driver or what?

It will be in the form of a BIOS update from the motherboard manufacturer.

[martin-w 3,235]

Stress Testing

 

Users should avoid running Prime95 small FFTs on 5960X CPUs when overclocked. Over 4.4GHz, the Prime software pulls 400W of power through the CPU. It is possible this can cause internal degradation of processor components.

 

 

 

Rob might be interested to know that the link he posted from Asus and the above statement, refers to the CPU automatically over volting when AVX instruction sets are used. This is Intel spec of course and not the fault of the Prime95 developer. However, it is of course right that Asus warn us of this occurrence when running small FFT's on a CPU overclocked above 4.4GHz. If the CPU behaves this way, and the user has already increased voltage markedly, then excess voltage and heat can indeed accelerate degradation of the CPU.

[Guest]

Here is a good read from the Intel forums regarding the discovery of this bug (before it was officially announced).  To be NOTED, Overclocking, normal, under-clocking conditions having no impact on the issue, it can happen regardless of clock speed with the correct series of complex instructions.

 

https://communities.intel.com/thread/96157?tstart=0

 

It's a somewhat humorous read with Intel finally admitting the problem.  I'd be interested to see what the publication contains when released and exactly how they "work around" this problem without a CPU design change ... since it doesn't appear to be clock related, I'd hazard a guess they use a less efficient set of instructions to do the processing (implications being performance penalty).

 

Cheers, Rob.

[GHD 21]

Mentioned in:

 

http://www.bbc.co.uk/news/technology-35361090

[martin-w 3,235]

People tend to think CPU's just work, and don't normally associate them with issues. So just for a laugh, check out the Haswell errata. Some 5 pages of flaws, most labelled as "no fix.
 
 
http://www.intel.com/content/dam/www/public/us/en/documents/specification-updates/4th-gen-core-family-desktop-specification-update.pdf
 
 

Sometimes bugs are disclosed, sometimes they aren’t — Piledriver has a significant problem with 256-bit AVX instructions, for example, that injects an 18-20 cycle delay into executing multiple consecutive instructions. Every original Intel Atom (before Bay Trail) had a floating point flaw that could insert a NOP (no operation) into every other cycle, effectively doubling FPU compute time. No one bought an Atom for its FPU performance, so the bug didn’t get talked about.

 

 

 

http://www.extremetech.com/computing/220953-skylake-bug-causes-intel-chips-to-freeze-in-complex-workloads

 

Interestingly, there was an Asus UEFI fix for my board just a couple of days ago. And it is labeled as " update microcode".

 

It seems this bug is sporadic. Some have encountered it and some haven't. Some after a few minutes, some after hours.

[TheFamilyMan 280]

I think a good part of Asus' stance on "potential issues" with Prime95 comes from the Haswell introduction of adaptive voltage control, which when enabled (its default) boosts voltage significantly when AVX or FMA instructions are executed.  If someone is overclocking with increased voltages and has adaptive voltage enabled, running Prime95 will spike voltages to dangerous levels.

 

For information only:  The latest rev of intel's LINPACK benchmark, which is really all that IBT is, uses FMA instructions (that are tailor made for super fast execution of BLAS equations which are at the core of LINPACK).  If you really want to see way serious heat generation, get the latest version from intel and run its batch script.  My "everyday" 4770k 4.5 OC will instantly spike over 95c, and all stock will hit above 70c. BTW I am not recommending this for anything other than personal entertainment.

[martin-w 3,235]

I think a good part of Asus' stance on "potential issues" with Prime95 comes from the Haswell introduction of adaptive voltage control, which when enabled (its default) boosts voltage significantly when AVX or FMA instructions are executed.  If someone is overclocking with increased voltages and has adaptive voltage enabled, running Prime95 will spike voltages to dangerous levels.

 

 

 

Yep, absolutely Gonemad.

 

I mentioned that 5 posts up.

 

"Thermal entertainment"!  I like it.  :smile:

[HiFlyer 9,358]

I think a good part of Asus' stance on "potential issues" with Prime95 comes from the Haswell introduction of adaptive voltage control, which when enabled (its default) boosts voltage significantly when AVX or FMA instructions are executed.  If someone is overclocking with increased voltages and has adaptive voltage enabled, running Prime95 will spike voltages to dangerous levels.

 

For information only:  The latest rev of intel's LINPACK benchmark, which is really all that IBT is, uses FMA instructions (that are tailor made for super fast execution of BLAS equations which are at the core of LINPACK).  If you really want to see way serious heat generation, get the latest version from intel and run its batch script.  My "everyday" 4770k 4.5 OC will instantly spike over 95c, and all stock will hit above 70c. BTW I am not recommending this for anything other than personal entertainment.

 

 

Yep, absolutely Gonemad.

 

I mentioned that 5 posts up.

 

"Thermal entertainment"!  I like it.  :smile:

 

Now that's something I find an unlikely concern. Overclockers absorb information like that with their mothers milk, and from there it's quickly disseminated widely on any number of sites. If you aren't taking account of such things then you probably shouldn't be out there aggressively overclocking in the first place.

 

The issue has been known for quite a while among the usual suspects, so I don't think intel should be concerned with large amounts of people blundering into trouble.

[martin-w 3,235]

Now that's something I find an unlikely concern. Overclockers absorb information like that with their mothers milk, and from there it's quickly disseminated widely on any number of sites. If you aren't taking account of such things then you probably shouldn't be out there aggressively overclocking in the first place.

 

The issue has been known for quite a while among the usual suspects, so I don't think intel should be concerned with large amounts of people blundering into trouble.

 

 

I think you've grasped the wrong end of the stick.

 

It's nothing to do with Intel. Gonemad and I were referring to Asus advising care when using Prime95 for stress testing for the reason stated above.

 

 If you aren't taking account of such things then you probably shouldn't be out there aggressively overclocking in the first place.

 

 

 

You say you regard it as an unlikely concern... except that there are a multitude of individuals out there with limited experience that attempt to overclock. Not everyone that overclocks is experienced, some totally inexperienced. Thus it's right that Asus should warn their customers of the potential risk. One of the places beginners receive information is from Asus themselves, their forums and articles.

 

As for Asus not favoring Prime 95 for stress testing and why...that was discussed earlier in the thread. It may be related to the adaptive voltage issue, it is related to the fact that Prime may not be validated for new architectures and it is related to Asus not favoring any synthetic stress test that bares no resemblance to how we use our PC's in everyday use. 

[HiFlyer 9,358]

I think you've grasped the wrong end of the stick.

 

It's nothing to do with Intel. Gonemad and I were referring to Asus advising care when using Prime95 for stress testing for the reason stated above.

 

 

 

You say you regard it as an unlikely concern... except that there are a multitude of individuals out there with limited experience that attempt to overclock. Not everyone that overclocks is experienced, some totally inexperienced. Thus it's right that Asus should warn their customers of the potential risk. One of the places beginners receive information is from Asus themselves, their forums and articles.

 

As for Asus not favoring Prime 95 for stress testing and why...that was discussed earlier in the thread. It may be related to the adaptive voltage issue, it is related to the fact that Prime may not be validated for new architectures and it is related to Asus not favoring any synthetic stress test that bares no resemblance to how we use our PC's in everyday use. 

 

My mistake, I typed Intel when I meant Asus, but the principal stands. If you're inexperienced, and want to overclock, then you'd be insane to just dive in and start changing things. You're going to search first for a reliable beginners guide, probably from Tom's hardware or some other generally well regarded source, and read a bit before messing around with your expensive investment.

 

It's the same as people who come here with questions and maybe lurk a bit before digging into the innards of FSX whilly-nilly.

 

Asus however, is going to be conservative, (which is to be expected) and that's why people tend to go to them last, because generally they are going to point you toward fairly mild overclocks using their recommended software, and then what's the point?

 

An inexperienced user is going to follow a guide from someone with a successful overclock, and if (s)he's smart, will probably look at several video and forum guides. It's easy to locate the ones done by experienced people aware of the issues.

 

If people were really out there blowing up processors in any numbers, it would be all over the internet immediately. (which it's not)

 

(Remember the time when overclockers proudly wore blown processors on chains around their necks?)   B)

 

If the software from Asus or whoever else is useful and effective, then it will gradually join the other suites of software that overclockers and others use. If not, then.... not. For now, people trust and use prime95, and luckily it caught the current problem, which just might not have been caught so quickly, or maybe at all, otherwise.

 

Now the main concern is if there will be a penalty for the "Fix", and if the potential penalty is large enough, will intel be forced to more aggressively address people's concerns.

[martin-w 3,235]

My mistake, I typed Intel when I meant Asus, but the principal stands. If you're inexperienced, and want to overclock, then you'd be insane to just dive in and start changing things.

 

It's not about "changing things" we are talking about an issue running synthetic stress tests with adaptive voltage. But of course, insane or not, a multitude do dive in and change things! And of course a multitude of those with only a modicum of experience do indeed run Prime95. I recall I did while being way less experienced than I am now.

 

A little knowledge is a dangerous thing as they say. I've conversed with individuals here on the Avsim forum quite a few times and in my role as moderator on Just Flight over the years, who dive in to overclocking with limited knowledge. Thus, as I said before, it's right that Asus should warn their customers of the risks involved. And given that overcloking is so simple these days, with sophistcated UEFI's that require just a few parametrs to be adjusted, it's even more likely that inexperienced users will overclock. And then we have Asus's excellent 5 way optimisation auto overclocking for example. Asus auto rules are now very good indeed. And after a customer has utilised 5 way optimisation [and remember they are likely to be inexperienced if using auto overcooking] they may well read that they should stress test with Prime95... thus, again, it's right that Asus should warn of potential issues.

 

 

If people were really out there blowing up processors in any numbers, it would be all over the internet immediately. (which it's not)

 

I don't see your logic. Know one is saying anyone is "blowing up CPU's". It's merely Asus warning of potential issues with synthetic stress tests, Absolutely no different to any other source advising of the potential adaptive voltage issue. If other s can advise about this in forums, then why do you think Asus can't?

 

Gonemad [i don't know his real name] and myself have speculated that this might be "one" of the reasons why Asus aren't to fond of synthetic stress tests and that may be the case. They may recognize that many less experienced overclockers may try to run Prime with adaptive voltage. Of course, as I said previously there are other reasons also why Asus aren't to fond of synthetic stress tests. 

 

 

For now, people trust and use prime95, and luckily it caught the current problem, which just might not have been caught so quickly, or maybe at all, otherwise.

 

 

Many do still use Prime95 yes, but many experienced overclokers have stopped using it. Personally I don't. I find it more efficient to use stress tests that are closer to the  software I run in everyday use. 

 

Actually, it's only in Prime 95 that the current bug has manifested itself. It may never have been a problem in other applications.