Intel is investigating game crashes on top-end Core i9 desktop CPUs
- Thread starter JournalBot
- Start date
Upvote
68
(81
/
-13)
Hard Thoughts
Ars Scholae Palatinae
Wonder if we’re at the point where silicon erosion is an actual problem here.
To the downvoters: this is a thing, I've seen the SEM images. It's one of the ways they set the limits for currents and voltages in the first place.
To the downvoters: this is a thing, I've seen the SEM images. It's one of the ways they set the limits for currents and voltages in the first place.
Last edited:
Upvote
5
(21
/
-16)
Palpatine would be proudIntel: Use these power limits on our chips
Motherboard makers:
Upvote
5
(15
/
-10)
So... more evidence these are literally the same chips with a different name, just binned better lol.
Upvote
64
(74
/
-10)
Wish the parameters would come down from the top rather than up from the bottom.
This is such a bad look for them where it feels like they and every manufacturer who uses their chips is chasing benchmarks to drive sales at the cost of stability for their customers.
This is such a bad look for them where it feels like they and every manufacturer who uses their chips is chasing benchmarks to drive sales at the cost of stability for their customers.
Upvote
70
(72
/
-2)
31% more power for 1-2% gains?!? I'm not really a fan of over-regulation, but that kind of waste is just gross. It's the PC equivalent of rolling coal, unless you're on a 100% renewable power plan you're wasting our future for a digital appendage substitute. It's not the world's most pressing waste of resources, but just ugh.
Upvote
167
(185
/
-18)
My i9-14900K was crashy crashy with machine checks all over the place until I went into the BIOS and backed off a bunch of settings that were over-aggressive but also the manufacturer's defaults. This wasn't observed in games, but in software compilation. Before I figured out that it was the BIOS authors messing with me, I changed the RAM, the motherboard, and the power supply with no benefits.
The 14900K is a fast CPU but these weird overclocks from the motherboard makers are a bad idea.
The 14900K is a fast CPU but these weird overclocks from the motherboard makers are a bad idea.
Upvote
111
(114
/
-3)
Mungus the Unhyphenated
Ars Tribunus Militum
Two things I've learned from overclocking modern processors, going all the way back to Skylake on the Intel side:
1) Intel's memory controller is very power-hungry. If you don't keep stable power to the memory controller, everything destabilizes. Focus on feeding the memory controller stable power within design limits for voltage and current. Then go back and tweak power settings for cores.
Just dialing everything up to 11 won't net the highest performance. Efficiency wins, often by getting better performance per clock cycle via stability improvements without having to crank up power limits. Technically, this is good advice for any CPU, but Intel CPU's and boards can be really touchy in some ways.
* Power stability on today's motherboards with very fast bus speeds everywhere can sometimes lose performance/stability when any "Spread Spectrum" power options are turned on. "Spread Spectrum" is a means of slightly varying power frequencies on the motherboard in order to more easily to meet EMI specifications. Unfortunately, it can also destabilize your OC in some cases when you need rock-solid power delivery. Tun it off and test. (Also, I've never run into EMI problems with it off...)
2) As you increase power limits, heat output on the motherboard from sources other than just the CPU goes up. Case airflow matters even when you're running big CPU liquid coolers. The rest of the motherboard plus the GPU need plenty of good airflow to stay cool. Adequate fans are a must. Mesh panels and open-design cases aren't always as pretty or quiet, but they're typically more efficient for cooling the entire system compared to all but the best-engineered glass "fish tank" cases. Whatever you choose, make sure you're keeping air moving over the VRM heatsinks and that it has somewhere to go. Heat buildup can mess with power delivery if it gets out of hand.
1) Intel's memory controller is very power-hungry. If you don't keep stable power to the memory controller, everything destabilizes. Focus on feeding the memory controller stable power within design limits for voltage and current. Then go back and tweak power settings for cores.
Just dialing everything up to 11 won't net the highest performance. Efficiency wins, often by getting better performance per clock cycle via stability improvements without having to crank up power limits. Technically, this is good advice for any CPU, but Intel CPU's and boards can be really touchy in some ways.
* Power stability on today's motherboards with very fast bus speeds everywhere can sometimes lose performance/stability when any "Spread Spectrum" power options are turned on. "Spread Spectrum" is a means of slightly varying power frequencies on the motherboard in order to more easily to meet EMI specifications. Unfortunately, it can also destabilize your OC in some cases when you need rock-solid power delivery. Tun it off and test. (Also, I've never run into EMI problems with it off...)
2) As you increase power limits, heat output on the motherboard from sources other than just the CPU goes up. Case airflow matters even when you're running big CPU liquid coolers. The rest of the motherboard plus the GPU need plenty of good airflow to stay cool. Adequate fans are a must. Mesh panels and open-design cases aren't always as pretty or quiet, but they're typically more efficient for cooling the entire system compared to all but the best-engineered glass "fish tank" cases. Whatever you choose, make sure you're keeping air moving over the VRM heatsinks and that it has somewhere to go. Heat buildup can mess with power delivery if it gets out of hand.
Upvote
69
(71
/
-2)
On my asus board, disabling "Asus Performance Enhancement 3.0" sets the Intel default power limits. For me, asus forces a choice to enable or disable that setting with every bios update and when I put the first CPU in, so at least it's not turning off the limits with no notice. FWIW, 253W stock limit on a 14700K and have not had issues, though it will power limit throttle in benchmarks.
Upvote
24
(24
/
0)
Mungus the Unhyphenated
Ars Tribunus Militum
I agree 100%. The "automatic" settings and overclocks, whether in UEFI/BIOS or via software in the OS, are usually inefficient and prone to having stability problems.
There's still no substitute for building up your configuration in the UEFI/BIOS by hand and testing, testing, testing until you have things where you want.
Upvote
19
(20
/
-1)
Motherboards are supposed to VRM throttle if the load demands more than they can supply, so unless there's some defects on the CPUs themselves (not stable at spec voltage, etc) it sounds like some vendors getting caught cutting corners on the voltage regulation. Will be interesting to see who is impacted and what the fix is.
Upvote
25
(26
/
-1)
"Epic suggests using a BIOS power setting called "Intel Fail Safe" on Asus, MSI, and Gigabyte motherboards...."
What about ASRock? Are these chips running on ASRock MBs immune to the problem, or is "Intel Fail Safe" not available on ASRocks?
What about ASRock? Are these chips running on ASRock MBs immune to the problem, or is "Intel Fail Safe" not available on ASRocks?
Upvote
8
(8
/
0)
This issue is addressed pretty clearly in this video.
View: https://www.youtube.com/watch?v=s43Auv8ub7w
View: https://www.youtube.com/watch?v=s43Auv8ub7w
Upvote
27
(29
/
-2)
Stop wasting all that power browsing Ars. Go outside. Hold your breath and stop wasting all that oxygen.
Upvote
11
(42
/
-31)
Unless your primary gaming computer is a Steam Deck or something I don't believe you.
Upvote
24
(35
/
-11)
Just want to point out that the exploding AM5 X3D processors late last year was, AFAIK, a result of board manufacturers exceeding AMD spec on SOC voltage.
These board manufacturers are unscrupulous.
These board manufacturers are unscrupulous.
Upvote
59
(59
/
0)
I hardly worry about overclocking anymore, as the gains nowadays just aren't as much as years past. The most I care about is making sure the memory has its boost enabled (XMP), and that's about it.
The days when you could push the "turbo" button and see clear effects from speed boosts are past us. I did get some laughs from those games when everything went bonkers fast when you pressed Turbo on.
The days when you could push the "turbo" button and see clear effects from speed boosts are past us. I did get some laughs from those games when everything went bonkers fast when you pressed Turbo on.
Upvote
36
(38
/
-2)
Upvote
10
(25
/
-15)
Yeah, it's a number that's hard to believe. Hell, playing FarCry 6 on an old 3770k paired with a 1070ti often pulls 500 watts.
Upvote
-8
(5
/
-13)
Turbo boost is often a race to idle mechanism where the total joules used is lowered by executing a task quickly and returning to a low power state.
Upvote
18
(19
/
-1)
Sounds a lot like a failed overclock. Mostly stable most of the time but crashing in certain situations. Not really too surprising since recent high end Intel chips seem to be clocked within an inch of their life anyway. Certainly, well beyond the efficient portions of the power vs frequency curve.
Upvote
12
(15
/
-3)
Most of my gaming these days is on my Deck. Does that make it my primary gaming computer? It's well under 45W.
Upvote
12
(16
/
-4)
I ended up giving my Ryzen 9 desktop to my nephew for Christmas since I barely used it for 6 months.
Upvote
3
(5
/
-2)
Race to idle doesn't always pay off. If you're pulling an extra 30% power to go 1% faster, it'll be more efficient to sit at a lower power state for a bit longer.
Massively power hungry systems can get a unit of work done using less total energy if all that power is translated to more work getting done per unit time. That does not describe what Intel has been doing. Intel has gone way outside of the efficiency bell curve and is willing to burn any amount of extra power for any miniscule improvement because "big number good."
Upvote
29
(31
/
-2)
I still hop on my very powerful gaming PC for some things, but since Balatro came out, I haven't touched it. That's dozens and dozens of hours of gaming at probably like 10W or something, including the display.
Upvote
1
(3
/
-2)
Thing is "gaming" can mean anything from "1080p and I'm fine with scaling back details" to "4K desktop and I will buy hardware to push 120Hz native in ALL the games, without using upscaling or 'pretend frames', and no I will not fiddel with settings". Heck, my now close to midrange Radeon 6950XT uses close to 300W. Add to this 4K monitors costing about the same as/not much more than 1440p monitors*. GPU requirements and their power draw have gone through the roof in the last couple of years. For the "serious" gamer a power supply below 1000W is madness.
*it's the reason I have a 4K monitor, it was on a sale and was much cheaper than equivalent 1440p models, and I thought I was kinda done with new games but it turned out I wasn't.
Upvote
19
(19
/
0)
Interesting to me is that also e.g. Core i9-14900K vs. Core i9-13900K there's only 2% overall boost.
Long gone are days when we could see like 50% jumps in performance. Intel seems to be crawling forward with same pace as general economy and GDPs are.
Upvote
3
(7
/
-4)
MedicinalGoat
Ars Centurion
Asus' initial response to that whole fiasco went beyond unscrupulous into actively malicious until enough tech media started calling them out. They juiced everything to unsafe levels then refused to warranty anything when it went up in flames while quietly deleting culprit BIOS revisions from their website.
As for the latest Intel power issues I'm feeling some deja vu from the Pentium III 1.13 GHz from the far flung year of 2000. In the crazed rush for BIGGEST NUMBER Intel released an unstable pile of garbage that had to be recalled. This is just more of the same.
Upvote
25
(28
/
-3)
I'm very aware of that, I was responding to the comment that basically said no chip should ever use turbo boost.
Upvote
8
(8
/
0)
In the US, if you run a 1KW system full throttle for a year, let's say 8 hours a day it will generate a bit less than 1,200 Kg of CO2 on average (with a wide range of course based on the power generation method used). That's equivalent to burning about 130 gallons of gasoline, or about half of the average annual gas usage of a person with a driver license in the US. In Hummer terms, it's about 1,500 miles driven at 12mpg.
Of course, this leaves out the fact that consumers are not asking for this inefficiency from processor or motherboard manufacturers. In my opinion, Intel is to blame for this huge drift in expectations of what's normal in TDP for a desktop processor. I have a 65W processor that goes to 85W on all core boost (benchmarks only, basically), and even that annoys me a bit. Don't get me started on GPUs, where Nvidia has even gone so far as requiring a new type of power connector to deliver enough power to their cards.
Fun aside: 1,800W wall draw is enough to trigger most residential breakers. I can't wait until using high end Intel and Nvidia means people will need to upgrade their electrical system to get those tiny performance gains.
Last edited:
Upvote
11
(20
/
-9)
The 13900K and 14900K are effectively the exact same CPU. They took a 13900K, boosted the clock a couple hundred MHz (3-6% boost depending on which clock you consider), and incremented the generation number. Since performance doesn't scale perfectly with clockspeed, a 1-2% increase in performance is reasonable.
Upvote
24
(24
/
0)
I usually prefer to stay at least one generation behind, and my needs are not really much. Going forward, cpu performance is going to be the least of what i would need and buying in on low end seems much more sensible. The products are verified, tested and cheap, and usually the last ones released on a platform.
I can't see any downsides.
I can't see any downsides.
Upvote
8
(8
/
0)
MedicinalGoat
Ars Centurion
The good news is if you bought Intel then being a "generation behind" actually means you're current generation.
Upvote
33
(35
/
-2)