Intel Skylake CPU flaw is making systems crash under heavy load.

[Enderman]

As previous posts mentioned, some people do use the CPUs for these intensive workloads.

the thing is that there are no "intensive workloads" as intensive as prime95

especially for consumers

so no, this is clearly rare issue which 0.0001% of people will encounter, and that is because they think that torturing their CPU with prime95 is a good idea

 

for all i care, the people who think prime95 is a good test can get all the "system hangs and unpredictable behavious" that they want

[Dabombinable]

the thing is that there are no "intensive workloads" as intensive as prime95

especially for consumers

so no, this is clearly rare issue which 0.0001% of people will encounter, and that is because they think that torturing their CPU with prime95 is a good idea

 

for all i care, the people who think prime95 is a good test can get all the "system hangs and unpredictable behavious" that they want

I get those when I don't make sure that my overclock is stable with Prime95-when stress testing you always use the toughest to remove any doubt as to the stability of the overclock. Once again your showing your naivety.

prime95 is already quite infamous for frying even hasswell cpu's if adaptive voltage is used although it can happen with any stress test.

I'll put it this way, the FMA2 causes my i7 4790K to pull 130W at stock....

[Prysin]

they dont upgrade to the most recent CPUs

they upgrade to stuff that is 1-2 years old

skylake has only been out for a few months

any company upgrading to something only a few months old is making a stupid decision

 

and no this is not an issue for anybody since clients will not be running prime 95, calm down

Prime95 is by far not the only software using complex AVX tasks.

[BrightCandle]

Prime95 is definitely not some special torture test. It uses algorithms available in a lot of other software. I genuinely have no idea where this came from, its like somewhere a time warp has happened and the history of this program and what it actually does has been erased and we are left with some how believing its this torture tool only and it does something magical that a CPU doesn't normally do. None of that is true its a completely reasonable and fairly normal program that does maths just like say a video encoder which remarkably does quite a lot of similar calculations. If your CPU doesn't work in Prime 95 and it crashes then its broken or your overclock is unstable. If you can't control the temperatures of the CPU with Prime 95 then your cooling solution is not sufficient. It really is that simple.

 

There seems to be this trend recently to somehow wave our hands like stability or thermally controlled are some how soft targets because most games don't fully utilise a CPU. The problem is then Crysis 3 comes out and those same people are on the forums complaining the game is unstable and its just that the game uses the CPU more fully than most. CPUs are designed to run 100% all day long and to do so on efficient algorithms. The idea that somehow some programs are torture and others are stress is wrong. If it causes problems on a CPU then its faulty because what Prime95 does (find large prime numbers) is a completely reasonable and normal thing to do with a CPU.

[Enderman]

Prime95 is by far not the only software using complex AVX tasks.

explain then why there is no other program that heats up a CPU as much as prime95

[WolfDeville]

i've been at a lan with a guy that had a 4-way sli of -that one nvidia card that had the twin-gpu with each gpu on a seperate pcb, and a heatsink in the middle that was basicly a fire hazard- and we did a furmark run for the lelz of it, and he said at the start "i dont have thermal protection so i'm gonna kill it at 95°c"

 

thats the reason furmark still exists

GTX 295? Quite an old card - When was this LAN party?

[manikyath]

GTX 295? Quite an old card - When was this LAN party?

its the card logan uses as his coaster i think, but he picked up two of the dual-gpu cards for like... single digits

 

and it was 1.5 years back.

[Dabombinable]

explain then why there is no other program that heats up a CPU as much as prime95

https://en.wikipedia.org/wiki/FMA_instruction_set

 

I think Enderman has me blocked. Someone please show that link to him-it might help educate him on Prime95 and the instruction it uses:

[Swndlr]

I just want a tech release without issues

I.E - Nvidia and the 3.5GB issue and now Intel with this

Granted this isnt a bad issue at all, but cmon, with all the money these guys have youd think they could release something flawlessly

[MageTank]

explain then why there is no other program that heats up a CPU as much as prime95

Linpack AVX2 actually gets hotter than 48k FFT Prime95. Your misinformation on Prime95 is starting to wear thin on me.

[Enderman]

Linpack AVX2 actually gets hotter than 48k FFT Prime95. Your misinformation on Prime95 is starting to wear thin on me.

Ah I see

So why exactly does a consumer with an i3 or i5 need to run these massively intensive number crunching programs...?

[MageTank]

Ah I see

So why exactly does a consumer with an i3 or i5 need to run these massively intensive number crunching programs...?

Prime95 was not made as a stress test. It is used as one, because AVX2 is very complicated math, and stresses the CPU quickly. It is also easy to customize the FFT lengths to specific groups to stress test faster. 1344k is what we use to see if our Vcore is stable. 512 small, 4096 large is what we use to test our IMC. 8k-48k is what we use to test our cooling solution. However, it's original purpose was NOT for any of this. It is used to find Mersenne Prime numbers.

 

A consumer with a locked i3 or i5 should not be using a stress test program anyways, unless they were testing to see if their PSU could handle the stress. Even then, Prime95 is not the best method to do that. 

 

Stress tests in general differ per person. I personally use Prime95 and Linpack both. I run Linpack for 10 passes (It is much quicker, and hotter) to see if my overclock is pseudo-stable (Means i can keep lowering voltage, or increasing BCLK) until it crashes. Once it crashes, i dial back, and give it a deep Prime95 test. Normally 27 hours of blend for full 100% stability. However, if my ram has already been proven stable in the past, i would just do 10 hours of 1344k for vcore. Since i survived the heat of Linpack AVX2, i don't need to do 48k FFT. 

 

While that is my personal method for testing stability, others prefer programs that won't be as stressful, or won't get as hot. For these programs, overclockers generally recommend doing them for longer. Cinebench for example. Most overclockers will suggest people run Cinebench for a very long period of time (24-48 hours) to ensure it is 100% stable, where as 8 hours of Prime95 blend is normally considered 99% stable for the most part. 

 

If you are asking why someone would use a non-overclocked Skylake chip on Prime95, the answer is simple. They used the program for its original purpose listed above, and not as a stress test. That is how this problem was discovered (Intel cited Prime95 as the source for the discovery) and now they are fixing it, as any program with AVX2 can also cause problems. 

[Enderman]

Prime95 was not made as a stress test. It is used as one, because AVX2 is very complicated math, and stresses the CPU quickly. It is also easy to customize the FFT lengths to specific groups to stress test faster. 1344k is what we use to see if our Vcore is stable. 512 small, 4096 large is what we use to test our IMC. 8k-48k is what we use to test our cooling solution. However, it's original purpose was NOT for any of this. It is used to find Mersenne Prime numbers.

 

A consumer with a locked i3 or i5 should not be using a stress test program anyways, unless they were testing to see if their PSU could handle the stress. Even then, Prime95 is not the best method to do that. 

 

Stress tests in general differ per person. I personally use Prime95 and Linpack both. I run Linpack for 10 passes (It is much quicker, and hotter) to see if my overclock is pseudo-stable (Means i can keep lowering voltage, or increasing BCLK) until it crashes. Once it crashes, i dial back, and give it a deep Prime95 test. Normally 27 hours of blend for full 100% stability. However, if my ram has already been proven stable in the past, i would just do 10 hours of 1344k for vcore. Since i survived the heat of Linpack AVX2, i don't need to do 48k FFT. 

 

While that is my personal method for testing stability, others prefer programs that won't be as stressful, or won't get as hot. For these programs, overclockers generally recommend doing them for longer. Cinebench for example. Most overclockers will suggest people run Cinebench for a very long period of time (24-48 hours) to ensure it is 100% stable, where as 8 hours of Prime95 blend is normally considered 99% stable for the most part. 

 

If you are asking why someone would use a non-overclocked Skylake chip on Prime95, the answer is simple. They used the program for its original purpose listed above, and not as a stress test. That is how this problem was discovered (Intel cited Prime95 as the source for the discovery) and now they are fixing it, as any program with AVX2 can also cause problems. 

And what other programs that are not linpack or prime use AVX2?

Programs that a regular consumer buying consumer CPUs would run?

Im pretty sure no video editing, games, 3d modeling, folding, or mining programs use AVX2...

 

I know this can be an issue for server CPUs, but it is extremely unlikely that ANY company upgrades their entire infrastructure a few months after a CPU is released...lol

Consumers, who are the majority of people who have bought skylake CPUs, have no reason to run linpack or prime other than for stress testing

And since there are other programs which can do (much safer) stress tests, there really is no reason why one MUST use prime95 or linpack on a daily basis

 

So in short, everything you have said has backed up my point that this is not any kind of dealbreaking issue for anyone here.

There is no reason to get so upset because "oh noes i cant run prime95 every day anymore because I love torturing my CPU"

[WolfDeville]

its the card logan uses as his coaster i think, but he picked up two of the dual-gpu cards for like... single digits

 

and it was 1.5 years back.

Yeah GTX 295. I wonder how they perform in todays games like BF4, Witcher 3, Shadow of Mordor etc..

[manikyath]

Yeah GTX 295. I wonder how they perform in todays games like BF4, Witcher 3, Shadow of Mordor etc..

oh it was a joke, but it was hilareous, and i had nice warm feet the entire lan party.

[MageTank]

And what other programs that are not linpack or prime use AVX2?

Programs that a regular consumer buying consumer CPUs would run?

Im pretty sure no video editing, games, 3d modeling, folding, or mining programs use AVX2...

 

I know this can be an issue for server CPUs, but it is extremely unlikely that ANY company upgrades their entire infrastructure a few months after a CPU is released...lol

Consumers, who are the majority of people who have bought skylake CPUs, have no reason to run linpack or prime other than for stress testing

And since there are other programs which can do (much safer) stress tests, there really is no reason why one MUST use prime95 or linpack on a daily basis

 

So in short, everything you have said has backed up my point that this is not any kind of dealbreaking issue for anyone here.

There is no reason to get so upset because "oh noes i cant run prime95 every day anymore because I love torturing my CPU"

No folding programs use AVX2? And you are "pretty sure" of it? GROMACS has been using AVX for as long as i can remember. It used AVX 128-FMA for the FX CPU's, AVX 256 for Sandy and Ivy, and AVX2 for Haswell, Broadwell, Skylake, etc. It also gets just as hot as Prime95, something you said was not possible before.

 

The problem in this thread is not Prime95, like you seem to think it is. The problem is AVX2 is breaking on Skylake, and needs to be fixed. AVX2 is NOT just used in prime95. Did you know emulators use AVX2? PCSX2 added AVX2 to its emulator renderer. I just showed you above where a folding core uses AVX. 

 

I see you out and about on this holy crusade against Prime95, a program that was NOT ORIGINALLY INTENDED FOR STRESS TESTS, but you say nothing of Linpack, which was specifically designed to stress supercomputers with endless looping math in AVX2. Intel uses Linpack for their XTU tests. Granted, Intel tamed theirs, and uses way less memory in their tests, but its still Linpack nonetheless. 

 

In short, this is NOT a Prime95 problem. This is an AVX problem. If you bothered to read the source, it said Prime95 brought it to Intel's attention. Failing to address this problem could result in problems for people that fold GROMACS on Skylake, use their CPU's for PCSX2 emulation, or other AVX2 based computation. It's a serious problem, and that is why Intel is taking it seriously. The more i speak, the more you ask me to provide more information and evidence towards your own claims. It's my turn. Do you actually know what AVX2 is, and what it is used for?

[Enderman]

The more i speak, the more you ask me to provide more information and evidence towards your own claims. It's my turn. Do you actually know what AVX2 is, and what it is used for?

yeah its right here

https://en.wikipedia.org/wiki/Advanced_Vector_Extensions

all the info you could want

 

doesnt say its supported on windows 10 though, so idk what thats about

[MageTank]

yeah its right here

https://en.wikipedia.org/wiki/Advanced_Vector_Extensions

all the info you could want

 

doesnt say its supported on windows 10 though, so idk what thats about

I did not ask if you knew where to find it. I asked you if you knew what it was. As in, do you understand what it is, what it is used for, and why this is not just a Prime95 problem like you seem to think it is. I've already met every request you made. I showed you a program that is hotter than Prime95. I showed you a game emulation program that uses AVX2, to show you it is used in gaming. I also showed you that it is used in folding. Are there any other areas you would like to challenge me on? I am certain i can cover all the bases if need be.

[Enderman]

I did not ask if you knew where to find it. I asked you if you knew what it was. As in, do you understand what it is, what it is used for, and why this is not just a Prime95 problem like you seem to think it is. I've already met every request you made. I showed you a program that is hotter than Prime95. I showed you a game emulation program that uses AVX2, to show you it is used in gaming. I also showed you that it is used in folding. Are there any other areas you would like to challenge me on? I am certain i can cover all the bases if need be.

What?

Who's challenging anything?

I think all i did was ask some questions...

[MageTank]

What?

Who's challenging anything?

I think all i did was ask some questions...

No, you made assertions. "No program will get as hot as Prime95". "No games or folding programs use AVX2". Those statements are simply not true. I can name several programs hotter than Prime95, if you really want to test heat. Plenty of other folding cores use AVX2, if you need to know all of them. As for games, AVX2 might not be widely used, but that is not the same as "no game ever uses it". 

 

What I am trying to say is, do not speak in absolutes if you are not 100% certain what you are saying is fact. This is why Prime95 is in the current situation it is in now. One person fried an absurdly high overclocked CPU with a lackluster cooling solution, word spreads on the net, and everyone becomes terrified. We need less of that, not more. 

[Enderman]

No, you made assertions. "No program will get as hot as Prime95". "No games or folding programs use AVX2". Those statements are simply not true. I can name several programs hotter than Prime95, if you really want to test heat. Plenty of other folding cores use AVX2, if you need to know all of them. As for games, AVX2 might not be widely used, but that is not the same as "no game ever uses it". 

 

What I am trying to say is, do not speak in absolutes if you are not 100% certain what you are saying is fact. This is why Prime95 is in the current situation it is in now. One person fried an absurdly high overclocked CPU with a lackluster cooling solution, word spreads on the net, and everyone becomes terrified. We need less of that, not more. 

Well i never heard of anyone having any issues with skylake before intel announced that there was an issue with it, so what does that say about the severity?

Clearly if this was a problem for consumers someone would have found out earlier by running prime95 or a game that uses AVX2, no?

Do you have a skylake CPU?

 

Ive seen people on the reddit thread about this say they ran prime95 with no problem...

[bcredeur97]

lol you know how you "fix" hardware features? you turn them off.

 

 

read something the other day about consumer chips being pathetic and having most of things turned off in the microcode after just a year because of all the crazy hardware issues.

[MageTank]

Well i never heard of anyone having any issues with skylake before intel announced that there was an issue with it, so what does that say about the severity?

Clearly if this was a problem for consumers someone would have found out earlier by running prime95 or a game that uses AVX2, no?

Do you have a skylake CPU?

 

Ive seen people on the reddit thread about this say they ran prime95 with no problem...

I own two Skylake CPU's. However, this problem does not apply to me, as my Pentium G4400 lacks AVX, and my 6600T loses it the moment i overclock. I did however run Prime95 blend for 8 hours to test this, and it did not crash.  I will run it for 27 hours this weekend, to test every FFT length and see if a specific one causes it, though i doubt it. Again, this is not a Prime95 issue, but instead an AVX2 issue within Skylake. Intel believes they can fix it with a bios update, but i am wary to accept a new bios atm, given my non-K overclocking. I'll test it though just to be certain.

 

Also, you assume the average consumer folds, emulates, and runs Prime95. I know this is only going to further your point, but my point is also bolstered by it. The point being, Prime95 is perfectly fine. 

[Enderman]

I own two Skylake CPU's. However, this problem does not apply to me, as my Pentium G4400 lacks AVX, and my 6600T loses it the moment i overclock. I did however run Prime95 blend for 8 hours to test this, and it did not crash.  I will run it for 27 hours this weekend, to test every FFT length and see if a specific one causes it, though i doubt it. Again, this is not a Prime95 issue, but instead an AVX2 issue within Skylake. Intel believes they can fix it with a bios update, but i am wary to accept a new bios atm, given my non-K overclocking. I'll test it though just to be certain.

 

Also, you assume the average consumer folds, emulates, and runs Prime95. I know this is only going to further your point, but my point is also bolstered by it. The point being, Prime95 is perfectly fine. 

ok, that's good youre running some tests

intel didnt say specifically that it would crash, just vaguely "instability and unexpected behaviour"

im curious to know what that actually means

lemme know if you notice anything when running those intensive AVX apps

[WolfDeville]

oh it was a joke, but it was hilareous, and i had nice warm feet the entire lan party.

Haha awesome