Skylake CPUs can freeze running Prime95

HazardO · 11 Jan 2016 at 12:58

http://www.theregister.co.uk/2016/01/11/math_bug_splatters_skylake_intel_working_on_fix/

TheRegister said:
Intel has confirmed it's pushing out a BIOS fix for a bug that can freeze its Skylake processors.

The good news is that the bug's triggered by complex workloads. It was turned up by prime number experts the Great Internet Mersenne Prime Search (GIMPS), who use Intel machines to identify and test new large prime numbers.

A few weeks ago, this post popped up on an Intel community forum. The poster said the bug “can freeze any system that has a Skylake processor”.

Since the bug has been reproduced on Windows and Linux systems, and on different motherboards, that post posited that it had to be down to the Intel processor.

In the case of the GIMPS, the freeze was observed on the Prime95 application. Prime95 uses Fast Fourier Transforms to multiply extremely large numbers, and a particular exponent size (14,942,209) leads to the crash after “minutes or hours”.

Last Thursday, on the same forum, Intel confirmed the issue, posting that “Intel has identified and released a fix and is working with external business partners to get the fix deployed through BIOS.”

It told posters on the forum to get in touch with their OEMs for the fix.

Mersenne primes are prime numbers which obey the formula Mn = 2n - 1* (the first Mersenne prime is 3, and the largest currently known is 257,885,161-1).

GIMPS founder George Woltman wrote the Prime95 software, and Wikipedia notes the group has turned up all of the Mersenne primes confirmed since 1996. ®

* Where "n" is a prime number

Takes me back to the Pentium PRO days, we don't see many CPU bugs which have a real impact these days. Will the BIOS fix which I assume is a microcode update for the CPU limit performance in certain operations I wonder?

pastymuncher · 11 Jan 2016 at 13:10

I bet the "fix" will incorporate something to stop bclk overclocking.

DragonQ · 11 Jan 2016 at 13:13

Isn't it a bit strange that this can be fixed via a BIOS update? I'm sure the last time there was such a fix it decreased performance because the only way to fix the bug was to disable a particular feature. Might've been an AMD issue, that one.

8 Pack · 11 Jan 2016 at 13:14

Prime 95 no one needs!!!

HazardO · 11 Jan 2016 at 13:19

DragonQ said:
Isn't it a bit strange that this can be fixed via a BIOS update? I'm sure the last time there was such a fix it decreased performance because the only way to fix the bug was to disable a particular feature. Might've been an AMD issue, that one.

A lot of CPU instructions are implemented in microcode on modern CPUs. BIOS patching of microcode instructions has been around for a long time now.

8 Pack said:
Prime 95 no one needs!!!

Unless they're looking for prime numbers

But there are other ways to stress test CPUs like you've said before.

pandem0nium · 11 Jan 2016 at 14:03

8 Pack said:
Prime 95 no one needs!!!

'No one' only includes just those who only run benchmarks or cpu-z validations.

It's a good indicator of overall stability if you're after stable overclocks for work or gaming.

tribz · 11 Jan 2016 at 14:53

Be interesting to see if non k overclocking gets 'fixed' also.

Disco_P · 11 Jan 2016 at 15:00

8 Pack said:
Prime 95 no one needs!!!

Except for the people who wrote the program, use the program and subsequently found this bug.

Not to mention the implications of this issue. Imagine a firm that's spent millions on a skylake based super computer for number crunching only to find out that it can't be used cause it will crash.

Your opinion on Prime95 as a stress testing tool has no bearing on this issue.

P4Clock · 11 Jan 2016 at 15:33

TBF i have noticed sticking/stuttering on both my Skylake CPU's.

i thought it was down to a bed setting on my system, but at stock the boys i3 will lock up for about 5 seconds after boot and will sometimes freeze for a second or 2 when i open folders. CPU performs fine is gaming, it like when it go's from idle to under load there is lag.
When the i3 is overclocked it fine, because you have to turn off all the power saving rubbish so it removes to lag problem.

mmj_uk · 11 Jan 2016 at 17:39

HazardO said:
Takes me back to the Pentium PRO days, we don't see many CPU bugs which have a real impact these days. Will the BIOS fix which I assume is a microcode update for the CPU limit performance in certain operations I wonder?

The first Phenom had a rather catastrophic bug, I think the BIOS fix for that impacted performance by 15% or so. Given that this is quite an isolated case I don't think the fix will have any real effect.

I didn't realise GIMPS/Prime95 had found every prime number since 1996.

Dave2150 · 11 Jan 2016 at 17:49

pandem0nium said:
'No one' only includes just those who only run benchmarks or cpu-z validations.

It's a good indicator of overall stability if you're after stable overclocks for work or gaming.

Do do realize that your talking to 8pack, one of (if not the best) overclocker in the world?

Prime95 is for people stuck in the dark ages, CPU benchmarking/testing wise. Useless since Haswell.

8 Pack · 11 Jan 2016 at 17:52

Dave2150 said:
Do do realize that your talking to 8pack, one of (if not the best) overclocker in the world?

Prime95 is for people stuck in the dark ages, CPU benchmarking/testing wise. Useless since Haswell.

With a name like Disco P it does suggest dark ages anyway! lol

Everyone is entitled to there own opinion. But for me Prime has very little use at all.

pandem0nium · 11 Jan 2016 at 18:30

Dave2150 said:
Do do realize that your talking to 8pack, one of (if not the best) overclocker in the world?

Prime95 is for people stuck in the dark ages, CPU benchmarking/testing wise. Useless since Haswell.

Yeah, I realise that but I don't have to share his opinion even if he is an overclocking legend.

Disco_P · 11 Jan 2016 at 18:39

Dave2150 said:
Do do realize that your talking to 8pack, one of (if not the best) overclocker in the world?

Prime95 is for people stuck in the dark ages, CPU benchmarking/testing wise. Useless since Haswell.

This has nothing to do with overclocking, benchmarking or testing. It seems skylake cannot do complex maths which is a serious issue.

Prime95 is very useful for people with Ivy Bridge and older which is a significant amount of users seeing as they are very powerful CPUs. Hardly "dark age" tech.

8 Pack said:
With a name like Disco P it does suggest dark ages anyway! lol

Everyone is entitled to there own opinion. But for me Prime has very little use at all.

My name has nothing to do with the point I am making. Seeing as, in your opinion, "no one needs" Prime95 what should they use to find Prime numbers?

pandem0nium · 11 Jan 2016 at 18:44

8 Pack said:
With a name like Disco P it does suggest dark ages anyway! lol

Everyone is entitled to there own opinion. But for me Prime has very little use at all.

I'm not disco P.

We all have our own standards for overclocking stability and preferred tools to use. No need to slander what has long been a highly regarded tool.

joeyjojo · 11 Jan 2016 at 18:51

Since it's a hardware bug the chances of it appearing only in Prime95 is about zero. All they've said is it appears under "complex workloads", which could be just about anything.

Very surprising Intel didn't spot this in design.

rossiboy001 · 11 Jan 2016 at 19:48

Sounds like another tick for a 5820k if someone is deciding which way to go.

zoomee · 12 Jan 2016 at 19:17

pandem0nium said:
'No one' only includes just those who only run benchmarks or cpu-z validations.

It's a good indicator of overall stability if you're after stable overclocks for work or gaming.

Agree 100%. It's a great stability indicator and I've been wondering why people never kicked up a fuss about prime95 not working with it since skylake was released. Some here are quick to accept the program wasn't 'compatible' - pure lolz.

I've had many a system in the past corrupt the OS, BIOS and all sorts when they using silly overclocks, prime95 is the best tool I know which can test a good combination of mem, cpu and memory controller all in one. I've yet to see a game or app fail after finishing 8 hours of prime95 testing successfully.

Now its an 'official' bug - well what do you know lol!

Couldn't care less if it "ONLY" fails in certain scenarios - if it was me I'd want 100% stability and not have to worry if my PC decides it needs to do something 'scientific' or 'intensive'.

zoomee · 12 Jan 2016 at 19:25

8 Pack said:
With a name like Disco P it does suggest dark ages anyway! lol

Everyone is entitled to there own opinion. But for me Prime has very little use at all.

Those pushing for benchmarks such as yourself bud I can understand not wanting to or needing to run Prime95. And yes I agree it can degrade systems too when used excessively.

HOWEVER - I'm one of those sticklers on stability and couldn't care less about how far the system can be pushed if it isn't 100% stable. Overclocking is resource intensive (timewise) so it's something I like to ensure is 100% stable before I carry on using a system and forget about overclocking it again.

Suicide runs for teh lose!

/runs away from 8pack to hide

Dave2150 · 12 Jan 2016 at 19:53

Disco_P said:
This has nothing to do with overclocking, benchmarking or testing. It seems skylake cannot do complex maths which is a serious issue.

I suggest reading the article again, the issue is already fixed, not that it was an issue, as it took someone who uses Prime95 as their job to identify the issue.

This wouldn't have affected 99.9999% of Skylake users, even if it wasn't fixed.

Though it is fixed, so it's a non issue.