Intel processors crashing Unreal engine games (and others)

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

tamz_msc

Diamond Member
Jan 5, 2017
3,760
3,581
136
Fans, fanboys, it's all the same i.e. not permitted. Attack the post not the poster.
Someone that builds PCs and codes in C has vastly more knowledge than the average PC user.
Shifting goalposts - how typical. First you replied suggesting that someone who can write code implementing some random data structure in C knows something or two about building PCs, when the reality is that the two aren't even remotely related.

Now, you just threw in a conjunction linking disparate stuff in an attempt to save face.
"Outliers", because they don't fit the Intel apologist(tm) handbook. Intel could enforce these limits on reviewers and board makers (including mandating PL1/PL2), but they don't. Hmm, what could Intel possibly gain by allowing their hardware to run out of spec for internet benchmarks? o_O
Outliers as in reviews that specifically show more than typical gains for either side - how convenient of you to ignore that part of my comment. Some fans of a particular team tend to show their true nature quicker than you can get your PC infected browsing the internet on IE 11 and Windows 7.
And...I wonder... is it the same idea that was behind their SPEC cheating?
Unrelated topic.
Or how about this one?
Also unrelated.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,760
3,581
136
But that's exactly what testing "at spec" means. It includes forcing Intel's default PL1 / PL2, whose wattages are lower than the unchecked furnaces Intel allows in benchmarks.
At spec means following specs according to Intel datasheets. Default means what is allowed as stock behaviour - and that includes PL1=PL2 typically with unlimited tau.

The two are not the same.
 

BFG10K

Lifer
Aug 14, 2000
22,694
2,923
126
Shifting goalposts - how typical. First you replied suggesting that someone who can write code implementing some random data structure in C knows something or two about building PCs, when the reality is that the two aren't even remotely related.
Where. Show me. My first mention of code:

I'd never heard of this setting until now and had to look it up. I know how to write a double-linked list in C using free() and malloc(). I bet most people in this forum can't do that, so I'm not a computer illiterate moron by any stretch of the imagination.

Please highlight the part where I said being able to code means you can build a PC. Put up or retract your false statement.

Some fans of a particular team tend to show their true nature quicker than you can get your PC infected browsing the internet on IE 11 and Windows 7.
Yep, absolutely. That's on full display right now with your posts. Poor, poor lil' Intel, the $181 billion corporation needs tamz_msc to defend m'lady's honor. o_O

At spec means following specs according to Intel datasheet
Intel 14900K specs:

Code:
Processor Base Power 125 W
Maximum Turbo Power  253 W

If you're saying those numbers are a lie then I'd agree with you, given Intel have been redefining what power consumption means for years through various mental and marketing gymnastics. If not then we need to enforce them in reviews. Uncapped power isn't the same as 253W, brah.
 
Last edited:
  • Like
Reactions: TESKATLIPOKA

Thunder 57

Platinum Member
Aug 19, 2007
2,669
3,786
136
I also read an article on gamersnexus.net (and am very glad to see written articles again) where a guy said the 4700k would beat the 7800X3D in performance and efficieny. It went terribly for Intel.


In a lot of these scenarios, without any controls, AMD is not only producing better gaming performance in raw framerate, but at a lower power consumption. Since these two are directly related in a formula, that gives AMD an efficiency advantage in games. Even when we lock the power or the FPS, generally speaking, AMD appears to hold an advantage in gaming.

Plenty more gaming examples there. Just take the loss and argue ARL on Intel 20A (or N3B) will be better. Though IMHO that looks unlikely.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,760
3,581
136
I'd never heard of this setting until now and had to look it up. I know how to write a double-linked list in C using free() and malloc(). I bet most people in this forum can't do that, so I'm not a computer illiterate moron by any stretch of the imagination.
Your second sentence is an assertion of your coding ability. Your third sentence's first part is a display of narcissistic behavior. The second part is where you imply that you're "not a computer illiterate moron" (owing to your previous statement), with the implication that having to look up what ICC_max is goes beyond your ability, despite you not being a computer illiterate.

So here you have yourself strongly indicating that someone of your calibre ought not to have to look up datasheets to figure out what safety limits are there in Intel CPUs, because of your coding abilities - when the truth is that one can be great at coding without having an iota of any clue about the thermal and electrical characteristics of a CPU.
 

BFG10K

Lifer
Aug 14, 2000
22,694
2,923
126
Your second sentence is an assertion of your coding ability. Your third sentence's first part is a display of narcissistic behavior. The second part is where you imply that you're "not a computer illiterate moron" (owing to your previous statement), with the implication that having to look up what ICC_max is goes beyond your ability, despite you not being a computer illiterate.
"Narcissistic", that's a big word. Did you google search that before or after you google searched ICC_max?

So here you have yourself strongly indicating that someone of your calibre ought not to have to look up datasheets to figure out what safety limits are there in Intel CPUs, because of your coding abilities - when the truth is that one can be great at coding without having an iota of any clue about the thermal and electrical characteristics of a CPU.
Answer the question: does 253W on Intel's spec sheet mean "uncapped power"?

Answer the question.
 
Jul 27, 2020
16,099
10,157
106
Every mobo should have three big buttons in the BIOS in the overclocking/tweaking section.

STOCK - This will keep your system performing rock solid.

PERFORMANCE - Relaxes some of the limits to let the CPU hit higher temperature and get extra voltage/current going through it to get a reasonable boost in performance. Fairly stable. Minor stability issues, if any.

OVERCLOCK / UNLEASH / UNLOCK - Forget stability. Whatever happens to you, you brought it upon yourself.

Going into Intel CPU spec sheets to figure out stable BIOS settings is pure insanity. No one should have to do that. No BIOS should let a user change a setting to a dangerous value without properly warning them of the consequences.

This is laziness on Intel's part which translates into laziness on the mobo vendor's part. Every person buying a DIY PC is not an overclocker or enthusiast. This reeks of a conspiracy where Intel wants average Joe to buy more expensive HP/Dell/Lenovo pre-built Gaming PCs because they are less likely to do things that would lead to warranty issues or their users suffering crashes and other stability woes.

INTEL MUST DO BETTER!
 

DrMrLordX

Lifer
Apr 27, 2000
21,606
10,800
136
I set PL1 and PL2 to 105W on my Z690 rig just for efficiency sake, but I had no idea ICC_max was even a configurable setting.

Once you have to change out of the box settings you're already past 90% of users, whether Intel or AMD...

Limiting power consumption via PL reduction probably causes the system to ignore ICC_max anyway. From what I can tell, all it does is set the current limit the CPU could draw if all the other failsafes against that behavior permit it. It looks like it only comes into play if your mobo settings effectively let your system boost to the moon, e.g. PL2=253W becomes more of a joke than an actual limit.

Tweaking ICC_max seems pointless when you can (and should) be tweaking PL values instead, unless you're an extreme overclocker that may WANT your CPU running up against a user-definable current limit. If that's you and you know what you're doing, have at it!
 

gdansk

Platinum Member
Feb 8, 2011
2,078
2,558
136
Knew this was happening before it even was a story.
Did you encounter it? I have not encountered it on any of my Alder Lake machines, though they are 12600K and 12100 so less likely to be operating at the edge even if the motherboard gives them too much leeway.
 

H433x0n

Senior member
Mar 15, 2023
873
937
96
Limiting power consumption via PL reduction probably causes the system to ignore ICC_max anyway. From what I can tell, all it does is set the current limit the CPU could draw if all the other failsafes against that behavior permit it. It looks like it only comes into play if your mobo settings effectively let your system boost to the moon, e.g. PL2=253W becomes more of a joke than an actual limit.

Tweaking ICC_max seems pointless when you can (and should) be tweaking PL values instead, unless you're an extreme overclocker that may WANT your CPU running up against a user-definable current limit. If that's you and you know what you're doing, have at it!
It’s the other way around really, the PL1/PL2 limits are pointless and you’re better off just using IccMax. I suppose You could ignore IccMax if you set your PL1/PL2 to a value like 200W but it’s suboptimal.

With my 14900K running IccMax=307A, Cinebench R23 maxes out at 228W, power virus apps like Prime95 or YCruncher barely pull more than 200W. Cinebench R15 will pull 253W since it’s not as heavy though.
 

Schmide

Diamond Member
Mar 7, 2002
5,586
718
126
I'd never heard of this setting until now and had to look it up. I know how to write a double-linked list in C using free() and malloc(). I bet most people in this forum can't do that, so I'm not a computer illiterate moron by any stretch of the imagination.

With standard library (std::list) and the emplace allocators of c++ only an illiterate moron would implement their own doubly linked list in c. Using free and malloc are equally arcane in modern structured containers. All joking aside, programming and hardware are different disciplines even if their spaces overlap.
 
  • Like
Reactions: Tlh97 and coercitiv

zir_blazer

Golden Member
Jun 6, 2013
1,163
405
136
ICCMAX is pretty much performing the role that PL4 used to, it is there to limit the peak transients. I would assume than there has to be a benefit by expressing it as Ampers instead of Watts, most likely than it would not be affected by any Voltage shenanigans like vDrop depending on load.
 
  • Like
Reactions: coercitiv

MrTeal

Diamond Member
Dec 7, 2003
3,568
1,694
136
410W power draw with a "spec" TDP of "150W", LMAO.

I'm not necessarily against Intel releasing a 400W CPU; there's definitely a market out there for the fastest thing possibly and damn the cost.
It's ridiculous that they still list it as a 150W TDP though. Anyone who's designing a thermal solution for a 14900KS and actually using that value is going to get no more performance out of it than a 14900K since both will be throttling at Tjmax. If you're going to do that, own it and list a reasonable TDP.
 

TheELF

Diamond Member
Dec 22, 2012
3,973
730
126
And so would their performance. Intel gets creamed at lower wattages.

Just saying, but the stated TDP has nothing much to do with how much power they actually draw.
We can see that intel draws 330 on a 253W power limit, which means that the power limit WAS NOT ENFORCED.
Also ZEN draws more power than intel at every other TDP limit below the highest one.
130462.png

If you limit the 14900k to 200W it will use 141W on average against the 128W average of the 7950x and will still be faster overall, zen is still more efficient just by far not as much as people think.
So no, intel does not need to run their CPUs at 350-400W to make benchmarks look better.
Intel could force a 200W limit and still be on top.
Gaming has worse efficiency because they don't park the e-cores, does zen still use game mode turning off one cxx?

D5TipA9.jpg
 

Thunder 57

Platinum Member
Aug 19, 2007
2,669
3,786
136
Just saying, but the stated TDP has nothing much to do with how much power they actually draw.
We can see that intel draws 330 on a 253W power limit, which means that the power limit WAS NOT ENFORCED.
Also ZEN draws more power than intel at every other TDP limit below the highest one.
130462.png

If you limit the 14900k to 200W it will use 141W on average against the 128W average of the 7950x and will still be faster overall, zen is still more efficient just by far not as much as people think.
So no, intel does not need to run their CPUs at 350-400W to make benchmarks look better.
Intel could force a 200W limit and still be on top.
Gaming has worse efficiency because they don't park the e-cores, does zen still use game mode turning off one cxx?

D5TipA9.jpg

You literally pick the one case where Intel looks better. All the rest, (except CB 1T, which is meaningless), AMD does better in. I can cherry pick too:

130513.png
 

BFG10K

Lifer
Aug 14, 2000
22,694
2,923
126
I'm not necessarily against Intel releasing a 400W CPU; there's definitely a market out there for the fastest thing possibly and damn the cost.
It isn't the fastest, though.

Threadripper is far faster at multi-threading and still uses less power.
X3D is faster overall in gaming and uses far less power.

It's ridiculous that they still list it as a 150W TDP though. Anyone who's designing a thermal solution for a 14900KS and actually using that value is going to get no more performance out of it than a 14900K since both will be throttling at Tjmax. If you're going to do that, own it and list a reasonable TDP.
Intel has been playing this game for years. Every CPU generation increases the gap between base and turbo, and between "TDP" and power draw @ wall. If someone points this out, they get told it's their fault for "not doing proper research". o_O
 
Last edited:

MrTeal

Diamond Member
Dec 7, 2003
3,568
1,694
136
It isn't the fastest, though.

Threadripper is far faster at multi-threading and still uses less power.
X3D is faster overall in gaming and uses far less power.


Intel has been playing this game for years. Every CPU generation increases the gap between base and turbo, and between "TDP" and power draw @ wall. If someone points this out, they get told it's their fault for "not doing proper research". o_O
\shrug It's the fastest they can make.
I wouldn't have an issue with AMD releasing an X3D chip either that had a higher frequency and power limit and let you pump hundreds of watts into it as long as your ridiculous cooling system kept the heatspreader at 50C. I'd just hope they wouldn't pump it out as a 7950X3D XT PE Ti Xtreme and still list it as a 120W chip.
 

DrMrLordX

Lifer
Apr 27, 2000
21,606
10,800
136
It’s the other way around really, the PL1/PL2 limits are pointless and you’re better off just using IccMax. I suppose You could ignore IccMax if you set your PL1/PL2 to a value like 200W but it’s suboptimal.

With my 14900K running IccMax=307A, Cinebench R23 maxes out at 228W, power virus apps like Prime95 or YCruncher barely pull more than 200W. Cinebench R15 will pull 253W since it’s not as heavy though.

You seem to be missing the fact that people are getting system instability from allowing ICC_Max to remain at 307a while the PL values are effectively being ignored by the motherboard.
 

H433x0n

Senior member
Mar 15, 2023
873
937
96
You seem to be missing the fact that people are getting system instability from allowing ICC_Max to remain at 307a while the PL values are effectively being ignored by the motherboard.
No, I'm not. You'll never get anywhere near 300W with IccMax=307A, it's the limiting factor 9 times out of 10.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,760
3,581
136
ICCMAX is pretty much performing the role that PL4 used to, it is there to limit the peak transients. I would assume than there has to be a benefit by expressing it as Ampers instead of Watts, most likely than it would not be affected by any Voltage shenanigans like vDrop depending on load.
Icc is the parameter that is used as a proxy for power draw in all IC design. Therefore Icc_MAX is what is specified at the design phase.

Icc and VID determine the actual voltage draw (Vcc) when the socket loadline impedance R and the VR's tolerance band (TOB) is known.

Intel's formula is the following:

Vcc_maxVcc = VID - Icc*R
Vcc_typVcc = VID - TOB - (Icc*R)
Vcc_minVcc = VID - 2*TOB - (Icc*R)

This is how it works out in practice, as shown for a Socket 478 Pentium 4:

1709365175024.png


Now for Raptor Lake, Intel's specifed DC_LL (R) is

1709365297404.png

And it is a fact that moterboard manufacturers tinker with this value, along with the others already discussed, to provide maximum performance out of the box, regardless of sample variation of the CPU.