NH-D15 can't handle an i9-10850k? What am I doing wrong?

somebrains

[H]ard|Gawd
Joined
Nov 10, 2013
Messages
1,425
It's not intel. There's a thread buried somewhere in the intel motherboard forum - it's the motherboard makers trying to make the review benchies look good, by being "faster" than the others.

IIRC, Asus doesn't do this - but everyone else does. Gigabyte and MSI pushed it the most, I think - playing with voltages and the like. My z490 Master had all-core max boost set from the factory. Have it off right now so I can sanity check a weird crash (bad driver from windows update, I think).

Asrock is pretty stable when you turn off the out of the box auto boost also.
 

lopoetve

Fully [H]
Joined
Oct 11, 2001
Messages
30,504
Asrock is pretty stable when you turn off the out of the box auto boost also.
All of them can be - it's just who's more aggressive :p That's partially why I bought the gigabyte though - less time to fiddle right now, and I'm open loop, so... not so worried about heat. ASRock is a bit more conservative though than the others.
 

zandor

2[H]4U
Joined
Dec 14, 2002
Messages
3,338
It's not intel. There's a thread buried somewhere in the intel motherboard forum - it's the motherboard makers trying to make the review benchies look good, by being "faster" than the others.
I was thinking about my chart rather than what the board makers are up to. I'm pretty sure the performance scaling based on power input is Intel's doing, though of course the board makers have to provide a way to set it and supply enough power, etc. and I had to supply a cooling system.

I get why the board makers turn off power limits, I just don't like it. It's like the twitchy gas pedals you get in some cars so they feel snappy during a test drive even though it's a midsize sedan or compact SUV with a 2.5l NA 4cyl or monitors being set to 9300k by default because it looks good under the fluorescent lighting in a retail store even though the color quality is crap. "Winning" at benchies when the reviewer doesn't know enough to check the power settings might be good for sales, but it's often not good for the customer.
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,209
And I quote, WITH the intel data sheet source:

The issue with Prime95 you "heard about" specifically refers to processors with "AVX" Instruction Sets, such as your 8700K. When any version of Prime95 with AVX code runs on AVX capable processors, it can impose an unrealistic 130% workload, which can adversely affect stability and severely overload your processor.

This is why BIOS has an "AVX Offset" (downclock) adjustment to compensate for the exceedingly high Core temperatures which are caused whenever the CPU encounters brutal AVX workloads imposed by certain utilities and scientific computational apps. Real-world software that uses AVX code, such as for rendering and transcoding, or some of the most recent games. do not impose extreme workloads.

Accordingly, as per Intel’s Datasheets, TDP and Thermal Specifications are validated “without AVX.

See page 87, Section 5.1.1, Thermal Considerations: 8th and 9th Generation Intel® Core™ Processor Families Datasheet, Volume 1 - https://www.intel.com/content/www/u...core/8th-gen-core-family-datasheet-vol-1.html


Note please the TDP and Thermal Specifications are validated without AVX portion.

I have an almost minor (dropped it my last semester years ago 2 credits short) in abstract mathematics, that at one point specialized in weather algorithms, chaotic dynamics, and HPC workloads for a major national scientific institution, before moving into a field that actually has money in it (my main degree is computer science, and I'm a systems architect). Bit rusty, but very familiar with what Prime and those guys do - the stress test is not necessarily the real-world workload, it's a stress test, and it's one that the processor was NOT designed for as per Intel.

This is all from the link in the second post.
I dont see that in any way or shape says what your claim was that "We're in a weird spot where a capability has been added to the CPU but it's not really designed to FULLY use it yet,"

Also again claiming a realworld application with a real world purpose is not realworld
if we go with the same excuse it means turbo boost is not something we shoudl trust. as intel measure TDP on base clock and turbo bost just like prime goes above the marketing value of TDP.

again to my oringla claim that just hidding instability by choosing not to test areas of your system that show to be unstable is really defeating the purpose of stress testing to begin with,

This truck is suppoed to haul 2 tons of rocks but it break down when it it hits 1.8tons.. fine we will just test it with 1.5 ton instead then.
if you can live with your instability or your truck only hauling 1.5tons of rock that is absolut your decision but dont come claim itsa 2 ton truck or a stable cpu
 
Joined
Nov 13, 2020
Messages
4
D14 & D15 user here.
"Higher values of LLC will generate an additional amount of heat"
That's targeted at us air cooling folk and can (almost) be disregarded completely by water cooling folk.
Lower your LLC and compensate with more vcore until stable (+0.010-0.030 ish).
Since i rock an asus board i can't give you a special LLC setting. Mine would be LLC5 from a range of 1-9 . 9 being the most LLC.

Usually it helps with the last "temp" hurdle.

And from one D15 user to antoher:
OC the hell out of it (5.1) but use adaptive vcore with the usual bios cpu power saving settings on and set VCCIO and System agent manually. Both have a sweet spot.
And use the P1 and P2 suggested constraints.

Good Luck !
 
Last edited:

Nenu

[H]ardened
Joined
Apr 28, 2007
Messages
19,458
D14 & D15 user here.
"Higher values of LLC will generate an additional amount of heat"
That's targeted at us air cooling folk and can (almost) be disregarded completely by water cooling folk.
Lower your LLC and compensate with more vcore until stable (+0.010-0.030 ish).
Since i rock an asus board i can't give you a special LLC setting. Mine would be LLC5 from a range of 1-9 . 9 being the most LLC.

Usually it helps with the last "temp" hurdle.

And from one D15 user to antoher:
OC the hell out of it (5.1) but use adaptive vcore with the usual bios cpu power saving settings on and set VCCIO and System agent manually. Both have a sweet spot.
And use the P1 and P2 suggested constraints.

Good Luck !
You are suggesting he remove the benefit of LLC (reduce vdroop during load) and instead increase voltage at both idle and load.
Not the best advice.
 

D-EJ915

[H]ard|Gawd
Joined
Jan 31, 2003
Messages
1,248
You are suggesting he remove the benefit of LLC (reduce vdroop during load) and instead increase voltage at both idle and load.
Not the best advice.
Eh not really, at load you get same voltage but you won't crash coming to idle from spikes like lots of people do with crazy LLC.
 

zandor

2[H]4U
Joined
Dec 14, 2002
Messages
3,338
I dont see that in any way or shape says what your claim was that "We're in a weird spot where a capability has been added to the CPU but it's not really designed to FULLY use it yet,"
I think I might have to make another chart using a processor I bought in 2012. It supports AVX, so we'll see what happens. Last time I OC'ed it was probably in 2012 and AVX wasn't really a thing yet. I'm not sure I'd even heard of it back then.
 

Nenu

[H]ardened
Joined
Apr 28, 2007
Messages
19,458
Eh not really, at load you get same voltage but you won't crash coming to idle from spikes like lots of people do with crazy LLC.
Fair play, I used the same method recently on an AMD 3600 build.
Its helpful to define the need for the method, his CPU might cope without boosting overall voltage.
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,209
Sigh.
Turbo is included in the TDP calculations. AVX is not. End of story.
lol. Love how you tried to get the last word in but not quoting because you were scared of a reply

https://www.intel.com/content/www/us/en/support/articles/000055611/processors.html
What is the maximum power consumption for my processor?
Under a steady workload at published frequency, it is TDP. However, during turbo or certain workload types such as Intel® Advanced Vector Extensions (Intel® AVX) it can exceed the maximum TDP but only for a limited time , or

  • Until the processor hits a thermal throttle temperature, or
  • Until the processor hits a power delivery limit.

You are indeed wrong according to Intel itself. they clearly state TDP is not to be seen as a maximum.
and not only can turbo and AVX exceed the quoted TDP but also other undefined workloads.

So yeah if your claim is that instability is excepted on loads that go above TDP that is not just for AVX but also Turbo boost and "Certain others loads"


Also why do you think Intel even put up the warning you were quoting if they though instability is just the norm (like your argument states) the fact that they DID encourage to ensure proper cooling for stability shows that the mentality for AVX is not "instability is OK" just that you need to be careful and AVOID instability

Again deciding to run on and unstable system is your choice. But trying to call it stable by ignoring instability and bringing up bad excuses on "stress testing beeing to hard", does not make it actual stable.


Not only are you factual incorrect on TDP here, but also nothing you quoted so far supported your original statement.
 
Last edited:

lopoetve

Fully [H]
Joined
Oct 11, 2001
Messages
30,504
So your quote, “until it hits a thermal or power limit” is exactly what is happening. I’m not sure why you’re having trouble understanding that? If you use AVX benchmarks, you will hit a thermal limit and the processor will throttle, at stock speeds, which is what this thread is all about- and that’s at stock speeds- again, not overclocked. Because that was not included in the default TDP, which by design is the limit the chip can pull under normal operations, therefore AVX is not part of the normal operations.

Look, you want to try and get a processor cool for AVX benchmarks, you do you, especially overclocked. Most of us have no need and aren’t going to bother, because I’m not using my system to run AVX benchmarks- especially it overclocked. If it passes standard workloads and stress tests without that, I’m fine- because those are part of the standard TDP design. AVX doesn’t do anything for me that I need, and no normal workload submits only AVX processes to the queue.
 

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,209
So your quote, “until it hits a thermal or power limit” is exactly what is happening. I’m not sure why you’re having trouble understanding that? If you use AVX benchmarks, you will hit a thermal limit and the processor will throttle, at stock speeds, which is what this thread is all about- and that’s at stock speeds- again, not overclocked. Because that was not included in the default TDP, which by design is the limit the chip can pull under normal operations, therefore AVX is not part of the normal operations.

Look, you want to try and get a processor cool for AVX benchmarks, you do you, especially overclocked. Most of us have no need and aren’t going to bother, because I’m not using my system to run AVX benchmarks- especially it overclocked. If it passes standard workloads and stress tests without that, I’m fine- because those are part of the standard TDP design. AVX doesn’t do anything for me that I need, and no normal workload submits only AVX processes to the queue.

lol is that what you learned form that
it states that it will go above TDP until you hit thermal threshold. aka start throttling which is correct as throttling lowers the physical load. that is true for ANY load with subpar cooling and has nothing to do with AVX.
the thermal threshold is based on temperature NOT TDP. You should know the difference
You can hit the thermal threshold way below the TPD by just removing the cooler.

If you go above the thermal treshold it simply means your cooling was sub par. not tha the load was to much and we should not cool for it.

Absolut NOTHING in that text support you claim that is the norm that AVX make you CPU throttle in normal speeds. You are really using a lot of imagination to bring up that claim/

I also love how you again complete ignores pats of the debate with your claim about turbo boost? and just tries to ignore those facts to make it fit an incorrect statement/

Your argument is based on a misconception
AVX is a perfect fine addendum to the CPU does it make the CPU hotter yes. are these functions meant to be run and used stable. yes
AVX is used in a lot of software since vector calculation are and actual thing we uses within math.

Bottom line is still. if you system cant run AVX overclocked, you system is unstable/undercooled. Just like the the Intel quote you gave.
and no amonut of you ignored facts whatever it is results form stability test or facts about TDP and turbo boost, is going to make it stable.


Heck if by you claim you CPU is hitting the Tjunktion point and acitation the #Prochot flag on clock speed with AVX. you must be pretty bad at building computers.
 
  • Like
Reactions: Mega6
like this

lopoetve

Fully [H]
Joined
Oct 11, 2001
Messages
30,504
Ok, lets rephrase this and set aside stability for a moment.

Do you believe that you can keep an Intel 9XXX/10XXX processor (setting aside HEDT and AMD implementations), running an AVX stress test, at sub-throttling temperatures, short of exotic cooling? Because intel does not believe so, nor does anyone else in the industry, as far as I can tell/test/see, which is why it is called out in the documentation. This thread is asking of the Noctua cooler is inadequate, and my answer to that is "No, it's fine - any AVX stress test will hit thermal throttle, because that's by design - the CPU cannot dissipate that kind of workload efficiently enough for any non-exotic cooler to keep you under the throttle limits, as the evidence shows." We've gone down rabbit holes of other parts, and some of it is potentially ambiguous, but the above statements should be as straight forward as they come.

edit: Or extremely high end water loops.

edit2: Or extremely high temps, rather than potentially throttle (again based on what your UEFI/Motherboard decides to do with things).
 
Last edited:

lopoetve

Fully [H]
Joined
Oct 11, 2001
Messages
30,504
And I'm not saying the full Tjunction temperature, necessarily - since part of that will be motherboard/UEFI controlled up till that point
 

chameleoneel

2[H]4U
Joined
Aug 15, 2005
Messages
3,752
Urr..

AVX/2 is used in a lot more than what you’re suggesting. It is not niche anymore.
Games typically make good use of AVX, as does office software and browsers.

One very common use is decompressing things into memory, or doing large memory fills from disk.

Most things can be trivially compiled to support it.

Many for loops and while loops can be vectorised with a little tinkering or just setting the compiler appropriately.

Thing is, AVX on intel is fast, faster clock for clock than AMD (at least on 3xxx silicon, not sure about 5xxx).

Here is a very simple example of where AVX can be used if the compiler flag is set:




Torvalds reckons NVidia is rubbish too.. thing is AVX can do a whole lot of calcs or compares at once. It is fast cause it is local in the core you are working with.

My next chip will have avx512 provided it is supported in a few generations.

The key thing is that the benefit of AVX is more significant than the offset difference in speed most of the time.

I run my 9900k at 5ghz without AVX offset, but those PL settings will rein it in and keep it quiet if it has a prolonged load. I lose 100-200mhz (varies depending on core) at most in those circumstances, while not thermally throttling and not running at stock (4.3-4.4ghz).

In light to medium AVX loads, there is no throttling, it’s 5ghz AVX all the way. I do have core parking and other power saving stuff enabled so if I’m only running on 4-6 cores those cores can sing...
it hasn't been said that AVX isn't used. It's been said there aren't common use cases, where the AVX capability is saturated.

yeah, some games use AVX. I've never had a game put my CPU over 90c. And that's because they do not flex AVX, like a prime torture test.


Also uh, I've had prime stable clocks still fail in handbrake and other stuff. So....I don't recommend leaning heavily on prime95.
 
Top