Linpack Xtreme Released

[Regeneration]

80C is this application? That's like 70-75C using something that isnt overkill. The amount of meat left on the bone here by using this app AND saying a max of 80C leaves enough meat on the bone for a Xmas meal.

Linpack and Prime95 (small FFTs) stress temperatures are in the same range. I was stating the maximum recommended temperature for most Intel CPUs is 80c.

Thank you to both, the machine goes in thermal throttling during the test, and throttlestop shows that some alarms are active, like PL1, PL2, VR Current and EDP OTHER. The Dell XPS is a hot pc.. so, the system provides to lower the temp around 84°C when it is throttling.
I know that every chip is different, but I've never heard of an i7 8750H which is unstable at -100mV

You should ask other i7 8750H users which stress testing software they used. Perhaps, none of them are stable with Linpack.

 

[Solid]

Undervolt is like overclock. A matter of luck.

The right way to run Linpack Xtreme: choose stress test, 10GB is sufficient enough, enter 10000 times and use all available threads.

For maximum stability, keep it running for 8-12 hours and ensure cooling is sufficient (20c below Tjunction, usually 80c).

So, do you think I should redo the test at -140 or -130mV using 10 GB for 12 hours or... since I've already got crashes/freezes using 30 GB, I should try with around -100mV?

 

[EarthDog]

Linpack and Prime95 (small FFTs) stress temperatures are in the same range. I was stating the maximum recommended temperature for most Intel CPUs is 80c.



You should ask other i7 8750H users which stress testing software they used. Perhaps, none of them are stable with Linpack.

his cpu throttles at 100C. Many people say keep it under 90c during stress testing. Saying 80c fo a 100c chip is too much meat on the bone....especially for an app that runs a bit warmer than p95.

 

[Regeneration]

his cpu throttles at 100C. Many people say keep it under 90c during stress testing. Saying 80c fo a 100c chip is too much meat on the bone.

Most desktop CPUs start to take damage and degrade slowly above 80c.

In addition, chance of getting errors increase by 30 percent regardless of voltage.

The maximum safe 24/7 temperature for most Intel desktop CPUs is 80c.

So, do you think I should redo the test at -140 or -130mV using 10 GB for 12 hours or... since I've already got crashes/freezes using 30 GB, I should try with around -100mV?

Run Linpack (bootable version) with a problem size of 35000 for 2-3 hours and be done with it.

If you passed Prime95 and other stress testing apps, it is possible the crashes are related to memory. Linpack is sensitive to memory errors.

 

[Solid]

his cpu throttles at 100C. Many people say keep it under 90c during stress testing. Saying 80c fo a 100c chip is too much meat on the bone.

yes the i7 8750H inside the Dell XPS 9570 starts to throttle around 97°C. At least on my machine.. but I think in every XPS.
If it throttles, the temps remain around 84, 85° C and the clock speed is reduced.
If it doesn't throttles, under load, the cpu clock is above 90°C, till 97°C and the clock speed is around 3,9 Ghz.
it is noway to test it maintaining the temp around 80° for 12 hours, unless it start to throttle all the time during test.

 

[Regeneration]

yes the i7 8750H inside the Dell XPS 9570 starts to throttle around 97°C. At least on my machine.. but I think in every XPS.
If it throttles, the temps remain around 84, 85° C and the clock speed is reduced.
If it doesn't throttles, under load, the cpu clock is above 90°C, till 97°C and the clock speed is around 3,9 Ghz.
it is noway to test it maintaining the temp around 80° for 12 hours, unless it start to throttle all the time during test.

Try to locate the laptop in a dry and cool place during the stress test. It is hard to ensure complete stability with the clock throttling.

 

[Solid]

Try to locate the laptop in a dry and cool place during the stress test. It is hard to ensure complete stability with the clock throttling.

Ok, thank you, i will try and I'll let you know.. also if I think that this cpu in a so thin case will go in thermal throttling anyway, and at any core voltage offset..

 

[er557]

-100mv is already a bit much and can generate whea errors and bsod in idle in the long run, even if passes quick stress test.
on another note, @Regeneration : you're saying even a server xeon with soldered ihs starts to degrade @89c linpack even though in real world it's nowhere near those temps-under 240mm AIO?

 

[Solid]

-100mv is already a bit much and can generate whea errors and bsod in idle in the long run, even if passes quick stress test.

Not for this cpu.. Online, every people who have this laptop were able to undervolt it of around -140 mV, someone of them also -170mV.
There are people who claim something about -180 or 200mV but I don't believe on them... anyway no one has problems at -100mV.
Also Notebookcheck undervolted this laptop of -115mV without any problems: https://www.notebookcheck.net/Dell-XPS-15-9570-15-more-performance-by-undervolting.317738.0.html

 

[Regeneration]

-100mv is already a bit much and can generate whea errors and bsod in idle in the long run, even if passes quick stress test.
on another note, @Regeneration : you're saying even a server xeon with soldered ihs starts to degrade @89c linpack even though in real world it's nowhere near those temps-under 240mm AIO?

Xeons and Mobile CPUs have reinforced silicon designed to survive in higher temperatures.

Here is an example:

i7-980X (Tcase 67.9c)

Same clocks, same architecture:

Xeon X5680 (Tcase 78.5c)

 

[EarthDog]

Most desktop CPUs start to take damage and degrade slowly above 80c.

In addition, chance of getting errors increase by 30 percent regardless of voltage.

The maximum safe 24/7 temperature for most Intel desktop CPUs is 80c.

Links supporting these assertions please. Sounds like bologna.

 

[er557]

on intel's own specs, tCase is usually 75c, and 80c is agreed to be a tad uncomfortable for nehalem architecture, which is what all latest years chips are based on.

 

[EarthDog]

100c tjmax on the chip in question.

Nehelam was over 8 years ago and its temps were a lot lower, yes.

 

[toyo]

on intel's own specs, tCase is usually 75c, and 80c is agreed to be a tad uncomfortable for nehalem architecture, which is what all latest years chips are based on.

That changed, now they use Tjunction. I think Intel knows what's best for their latest CPUs. der8auer even ran a 8700K without ANY cooling at all, and the PC didn't shut down or anything, it just ran at minimal speeds and throttled as necessary when reaching Tjunction=100C .

There's also the thing that laptop CPUs all reach these temps very fast under more stressing workloads and Intel doesn't make them throttle at all even at 95C. Same is true for my 8700K. It will run in full speeds, only when it reaches 100C it starts some throttling. If Intel knew that degradation occurs at 90C, you can bet they would have takes steps to make the CPU throttle faster.

 

[Solid]

There is the "Prochot function" which throttles the cpu if it is close to 100°C, to reduce the temp.

 

[EarthDog]

https://ark.intel.com/products/134906/Intel-Core-i7-8750H-Processor-9M-Cache-up-to-4-10-GHz-

TJMax is 100C. The CPU throttles around that point to protect itself. It has another measure later that shuts it down in throttling isn't doing it. Running that CPU at 80C during THIS test (and others like P95) is and saying 80C leaves a tremendous amount of meat left on the bone. Since few loads come close to This app and P95, That means temps are going to be on the order of some 10-20C lower if not more (read: gaming will be close to 30C less depending on the title).

If Intel knew that degradation occurs at 90C, you can bet they would have takes steps to make the CPU throttle faster.

Exactly.

Now, I am sure a CPU degrades faster the warmer it runs, but, we are talking about a CPUs useful lifespan here it will be ok. I've run anything from P4 to Coffee-Lake and stayed ~10C under TJmax and all CPUs have led a long life. Hell, my C2Ds and C2Qs are still going strong at those were pegged temp wise and highly overclocked and over volted.

The bottom line is 80C is an ultra conservative temperature limit for stress testing (this CPU), and especially this type of overkill testing application. If you don't want your CPU to throttle, don't run this test... don't run P95. I doubt the laptop's cooling can keep up. Run AIDA64 stress test. If you want to push things through that, run CPU and FPU only to get things cooking.

 

[toyo]

By the way, for the curious, you can see a 8700K without any cooler running here. Use captions for En subs. It will even run multithreaded Cinebench.

 

[Regeneration]

Links supporting these assertions please. Sounds like bologna.

https://www.intel.com/content/dam/w...orkstation-overclocking-risks-white-paper.pdf

See "Frequency Degradation" for a brief summary.

And to be a little more technical, let's pick a random CPU datasheet:

https://www.intel.com/content/dam/w...heets/8th-gen-core-family-datasheet-vol-1.pdf

"Operating the processor outside the specified limits may result in permanent damage."

The limits Intel is referring to is Tcase for lidded parts, Tjunction for bare die parts and according to TDP.

But let's make it easier for everyone to check for themselves:

If you stick your nose close to a desktop Intel CPU when its operating at 80c and above, you will smell something funny, and it gets worse with the temperature.

 

[EarthDog]

Thank you for the links.. if I may......

See "Frequency Degradation" for a brief summary.

1. Nobody is overclocking.
2. We all know that overclocking will reduce the lifespan of a CPU.
3. This warning makes complete sense for enterprise solutions (who the article is for) where data integrity is key more than speed. On the consumer side, Intel offers a separate warranty which covers overclocking.


Tcase - How are we measuring this? Does HWmonitor (is it the package?). Realtemp..nope. Coretemp... nope. Do any others (this is some of the more common ones)?. How do you propose an average user finds this temperature? Coretemp even says TJmax as it reads the cores. Why would this long standing application say such a thing if TJmax doesn't matter and Tcase is what should matter? Nothing really monitors Tcase... so how important is it really when no(?) temp program monitors this value nor does Intel's throttling seem based on it, but responds of TJmax temps. If people kept their temps to Tcase and below, lol, this stress test and others would be impossible to run without breaking that value. Too much misplaced importance on Tcase here it seems.

Are you able to explain why, if Tcase is so important, that the CPUs throttle using TJmax values regardless if has an IHS or not? One would think if a worrisome amount of degradation where to occur above Tcase, Intel themselves would use that value rather than base it off of something nearly 30C higher. Intel papers on his chip say Tjmax.. 8700k Tjmax... 9900k tjmax... please explain why everything is based on TJmax if that temp is so bad...

If you stick your nose close to a desktop Intel CPU when its operating at 80c and above, you will smell something funny.

LOL, Not true at all. I have benchmarked like this for decades and haven't smelled anything and ran at their max (below TJmax). This is bolgona.

"Operating the processor outside the specified limits may result in permanent damage."

Of course this is true. Since we are under TJmax, which is the temperature at which modern intel CPUs throttle, There isn't a worry. That passage also considers voltage and frequency changes as well.

 

[er557]

I would be uncomfortable pushing my hardware close to tjMax, also it brings few benefits if at all.
Long term safety and longevity is far more important, more than a few more points in a benchmark. Cpu and gpu likewise

 

[Solid]

mmh a question for @Regeneration: How can I test the stability from portable version (linux distro) if i can't start Throttlestop to undervolt the cpu from there? Can I use the normal version for windows instead? If yes i think i should go with 10 GB of Ram for at least 2 hours, right?

 

[EarthDog]

I would be uncomfortable pushing my hardware close to tjMax, also it brings few benefits if at all.
Long term safety and longevity is far more important, more than a few more points in a benchmark. Cpu and gpu likewise

You shouldn't be...

If we use this stress testing application and use Tcase as a max, that means CPUs may not be able to run stock without breaking that value. Same thing w/P95 (using AVX.FMA etc)...

Long term safety and longevity is far more important, more than a few more points in a benchmark. Cpu and gpu likewise

I agree, but we aren't talking about the CPU lasting for a couple of years because of this. The CPU will still last through its usefull life. Also, you have to consider that most stress tests, and this one in particular, runs things 20-30C warmer in most situations. It is rare one will ever be able to reproduce the temps we see with most activities. Gaming can be 40C cooler than this stress test. I can run this thing and hit 90C and when I game after 4 hours, I peaked at 58C. I encoded something for over an hour... 65C. Rendered (blender) around there.

I guess what I am saying is if we followed Tcase, which it seems one piece of software monitors (and calls it by the name of package), People couldn't overclock anything.

I can't tell you how many CPUs over the last 20 years I have run at the a very edge and those same CPUs still ran through its useful lifespan. That isn't to say none died, but, hell, I had chips die at stock speeds too...


EDIT: Bone stock on a 9900K with a 2x120mm CLC cooler I'm already at 71C after like 2 mins... once this thing saturates, I'm well past a 72C Tcase value...

EDIT2: 76C and counting after 4 mins using this...


Should I stop the test? 76C -5C is Tcase...right?

 

[Regeneration]

mmh a question for @Regeneration: How can I test the stability from portable version (linux distro) if i can't start Throttlestop to undervolt the cpu from there? Can I use the normal version for windows instead? If yes i think i should go with 10 GB of Ram for at least 2 hours, right?

Sorry, forgot about ThrottleStop. Yes, just run the Windows, for a few hours. No need to fry the laptop for 12 hours.

I can't tell you how many CPUs over the last 20 years I have run at the a very edge and those same CPUs still ran through its useful lifespan. That isn't to say none died, but, hell, I had chips die at stock speeds too...

You don't know that, because you don't run 'overkill' stress testing software to test stability and integrity.

So unless the CPU is completely dead, there is no way for you to know.

Most of my CPUs hit 70-75c with Linpack.

 

[MrGenius]

Tcase/Tcmax is a joke. Like if you can keep the lid under a certain temperature, the die can be run at any temperature(even well in excess of Tjmax). It's nonsense. The only thing that matters is Tjunction/Tjmax.

...Tcase, which it seems one piece of software monitors (and calls it by the name of package),...

It doesn't though. Unless you place a thermocouple directly on the lid of the processor you are not able to measure Tcase. Tcase = the temperature of the lid. Which is pretty much irrelevant as far as determining the temperature of the CPU die. Like I said, as if keeping the lid cool enough means you can run the CPU die hot enough to melt without issues. Tcase assumes way too much and is a stupid specification. I suggest everybody do like I do and completely ignore it.

 

[EarthDog]

You don't know that, because you don't run 'overkill' stress testing software to test stability and integrity.

So unless the CPU is completely dead, there is no way for you to know.

Most of my CPUs hit 70-75c with Linpack.

Umm. Yes I do know that they still ran. They still ran F@H on the CPU, they still rendered and encoded... The stability and integrity was there for my uses, which was pretty heavy on the CPU at times (running weeks on end F@H for example or other distributed projects). Remember, stability is in the eye of the beholder. If the system doesn't crash/freeze/slow up (low benchmarks due to instability) for how a PC is used, then it is stable for that user.


EDIT: Regen - Can you answer some of my questions, please? Why do you place so much importance on this value when seemingly nothing else (talking temp/monitoring programs) do? What do you know that all these others seem not to?

It doesn't though. Unless you place a thermocouple directly on the lid of the processor you are not able to measure Tcase. Tcase = the temperature of the lid. Which is pretty much irrelevant as far as determining the temperature of the CPU die. Like I said, as if keeping the lid cool enough means you can run the CPU die hot enough to melt without issues. Tcase assumes way too much and is a stupid specification. I suggest everybody do like I do and completely ignore it.

That is what I thought as well... but wasn't certain.

Either way, Tcase isn't even listed on processors I looked up (8750H, 8700K, 9900K just now - I know on some older processors it was, like Nehalem nearly a decade ago), it is only TJmax. The importance placed on this seemingly undefined (on modern processors) and elusive to measure value, seems waaaaaaaaaaaaay too high here.




I guess in the end is (my) fear of spreading misinformation to users. As with anything, running it out of specification is likely to reduce the lifespan of the item being run out of spec. It is why specs are there. What is important is to know the RIGHT specs to look at and what weight/relevance the specification is. The concern I have with Tcase and running his CPU under 80C is that I'm almost sure a laptop won't be able to do that stock. NOt to mention, that value seems not defined or tracked with any major temp program.


Note... after ~15 min test using Linpack (10GB of 16GB RAM use all c/t), I hit 78C with this CPU at stock. According to these 'rules' I would should not overclock at all as I will break 80C which causes degradation significant enough to warrant a warning about. I'm sorry... but that is just hooey. According to 'my' rules, that leaves 10C headroom to overclock... which is not a lot. This program is too much for the average user to do long term OC stability testing. If some prosumer number cruncher came through and needs 99.9999999999999999999% stabilty instead of 99.999999999999999999% stability I'd say, yeah, use this... pound it. But its overkill for the average user leaving too much meat on the bone. IMO, its usefulness for the average joe overclocker is perhaps with maybe 5 min runs......enough to get an idea of max temps (and then some) and see if its 'stable enough' to push forward one notch. Quick and dirty testing for the overclocker.


Regen - what is a comparable setting in your app to say Small FFT and blend in P95?