Intel CPUs Are Crashing and It's Intel's Fault: Intel Baseline Profile Benchmark

[Fastturtle]

As a "Gamer-Troublemaker", whatever the hell that means. Maybe Intel should do a better job at making sure that MB manufactures, and the public understand the power limits of there CPUs.
And by the way, why are Gamers trouble makers?
Because they discover a problem with a product?
If it wasn't for gamers pushing the hardware to the limits, you would still be running 486 machines
So instead of knocking people down, you should be thanking them.
Cause I don't see you contributing in any meaningful way other than trolling the subject matter

Wrong about Gamers and 486 chips, We'd still be using 286 and happy to have 4MB of memory while running Win 3.11

 

[WhoCaresAnymore]

Does this also affect the laptop CPU’s? Ones used in the micro pc’s?

 

[pmshah]

Ever since they delicenced VIA as "chipset" source their own products have sucked big time. Repeatedly crashing SATA drives and corrupting of USB storage were the norm. Even bigger issue was their own motherboards. One example was not being able to boot the system if I had front panel USB-3 connector plugged into the on board 19 pin header. It took more than a year and 2 bios updates to fix that. I remember how they tried to screw Zotac by rearranging onboard functions of Atom chipsets when Zotac used nVidia graphic chip on their dual core Atom processor to build a really good performance mini-itx motherboard.

 

[hwertz]

Does this also affect the laptop CPU’s? Ones used in the micro pc’s?

Technically yes, but as a practical matter no. This isn't a chip defect, it's Intel "suggesting" a power profile with a 253W cap, but not actually listing a maximum limit, so these enthusiast boards were having it suck down like 280-350W.

I hope there's no laptop on the market where it already had a 253W CPU, and they decided to have it draw 300 instead. That'd run a tad toasty. They do make a few models of smaller PCs with robust cooling so it's probably technically possible; but at least in the past this was so you could put a decent video card in there without the box melting down, not so they'd run the CPU at "enthusiast" speeds.

Re: the comment about gamer troublemakers... as a few said they haven't read the article. These board makers are not violating Intel's specs, Intel just has specs that are entirely too loose. On the other hand, I overclocked back in the day. And I find it troubling that the current push for the highest benchmark scores would lead vendors for, lets face it, gaming motherboards, to think it's appropriate to have a single "turn on XMP" have these CPU run at almost a 50% increase in power consumption and 50% increase in clock speed. I do think the option should be there, but it should be in with the overclocking optons where it's clear you're pushing things. Back to Intel, given motherboard makers can do this while still following your specs, your specs are too loose!

 

[daffy duck]

For further proof of what a fluster click this is turning out to be watch Buildzoid's (Actually Hardcore Overclocking) YT video on gigabytes Aorus Extreme Z790 with the new so-called stable defaults. It's a bit rambling but it shows the state of affairs is crazy. I'm going to watch his new video for Asus's Z790 mb as it uses different Intel defaults. Amazing how everyone has a different interpretation of what default means.

Please note Intel's claim 400A should be the max and is stable is total BS. The Intel spec sheet for 125W+ cpu's says the limit is 307A. 400A won't even allow Cinebench R15 to run without crashing on 14900K.

 

[hwertz]

Yeah man, I had an ECS motherboard back when they got caught using fake cache chips in some boards too. Luckily mine was fine (real cache, and it was a nice motherboard -- just an exact ripoff/clone of some other companies Via chipset based motherboard.) I heard those Via chips were bad in Windows but I was running Linux, on bootup it'd print notices about several Via chipset bugs it was working around; apparently successfully, that thing ran flawlessly.

As a PC technician who builds Rigs for my clients I've been using AMD in most of my builds in the last years

Yeah. You can run into this problem with AMD systems, but you have to have a BIOS that supports overclocking (or those power adjustment tools for Windows or Linux) and just start indiscriminately cranking up the juice. And they give you a warning about what you are doing.

As for your thoughts regarding Intel versus AMD... yeah. Back in the day I had a K5, K6, K6-2, Duron, Athlon XP. Running AMD then was a no-brainer... AMD and Intel took turns taking the actual performance crown, but (using current i3/i5/i7/i9 terminology) when AMD wasn't the fastest you could still get an i7-class CPU for like i3 to low-end i5 prices (plus, as now, AMD was big on reusing sockets, the whole time I went with AMD it was socket 7 then socket A). Around the time the Athlon XP came out, they were pretty even; Intel CPUs were better for a good while after that. It was with the Ryzen where AMD got their wind back and started making really nice CPUs again.

 

[letsgoiowa]

This seems like an astoundingly easy solution that Intel and board partners totally chose to ignore: PL stepping should be fixed and universal across all boards. If you want to change it, that's overclocking and SHOULD NEVER be default behavior! The user should have to go into UEFI or use XTU to enable that manually. The capacity to dynamically boost higher on its own is great...if it's opted into.
Intel is responsible for not requiring that (insecurity and greed). Motherboard manufacturers are responsible for not understanding the limits of their own engineering (hilarious incompetence).

 

[hwertz]

This seems like an astoundingly easy solution that Intel and board partners totally chose to ignore: PL stepping should be fixed and universal across all boards. If you want to change it, that's overclocking and SHOULD NEVER be default behavior! The user should have to go into UEFI or use XTU to enable that manually. The capacity to dynamically boost higher on its own is great...if it's opted into.
Intel is responsible for not requiring that (insecurity and greed). Motherboard manufacturers are responsible for not understanding the limits of their own engineering (hilarious incompetence).

Agreed. That's the way it always was in the past? Back in the day, some systems even required jumpers (to set FSB and multiplier); when they moved to having the settings in the BIOS, clearing the settings ALWAYS reset it to a safe speed.

Honestly, I think what's happened here is that the top of the line Ryzens are very fast, and Intel wanted some benchmarks where their 13900K/14900K were cleaning the Ryzen's clock. And assumed when these boards proved unstable, they'd be like "Oh, Gigabyte/MSI, come on, really" and let people assume these boards were running the CPU out of spec, they'd release a BIOS update that tamed the speeds a bit but the high benchmarks would already be out there. I don't think they counted on Gigabyte (and then the tech sites) picking up on these boards WERE following Intel's specs, the specs are just way too loose.

I do agree that these motherboard makers definitely share responsibility here -- Intel specs or not, Cinebench will crash on these things in like 15-60 seconds. It does seem like testing a new motherboard with the top-of-the-line CPU (and a large cooler so it doesn't thermal throttle) would be a good idea; I mean like 5 minutes of testing would have shown them there's a problem.

But mostly it's on Intel, as you say they should have a pretty strict base spec, and going above that spec should be a matter of clearly turning on overclocking settings. (AMD and Intel CPUs usually have lower TDP targets settable too, for notebooks to keep power under control and prebuilts where they know the cooling is trash... well, that's for small form factor systems and embedded really. But hopefully one would buy a less expensive CPU rather than buy a 14900K or top-of-the-line Ryzen then kneecap it with a reduced TDP limit.)

 

[Angga B]

Since we knew that there is a common process in chip production called as "binning" then we should also knew straight away that this is merely an utter example of GREED. In order to ship as many top of the line products they simply relax the binning parameters. Period.

 

[KaitouX]

Agreed. That's the way it always was in the past? Back in the day, some systems even required jumpers (to set FSB and multiplier); when they moved to having the settings in the BIOS, clearing the settings ALWAYS reset it to a safe speed.

Honestly, I think what's happened here is that the top of the line Ryzens are very fast, and Intel wanted some benchmarks where their 13900K/14900K were cleaning the Ryzen's clock. And assumed when these boards proved unstable, they'd be like "Oh, Gigabyte/MSI, come on, really" and let people assume these boards were running the CPU out of spec, they'd release a BIOS update that tamed the speeds a bit but the high benchmarks would already be out there. I don't think they counted on Gigabyte (and then the tech sites) picking up on these boards WERE following Intel's specs, the specs are just way too loose.

I do agree that these motherboard makers definitely share responsibility here -- Intel specs or not, Cinebench will crash on these things in like 15-60 seconds. It does seem like testing a new motherboard with the top-of-the-line CPU (and a large cooler so it doesn't thermal throttle) would be a good idea; I mean like 5 minutes of testing would have shown them there's a problem.

But mostly it's on Intel, as you say they should have a pretty strict base spec, and going above that spec should be a matter of clearly turning on overclocking settings. (AMD and Intel CPUs usually have lower TDP targets settable too, for notebooks to keep power under control and prebuilts where they know the cooling is trash... well, that's for small form factor systems and embedded really. But hopefully one would buy a less expensive CPU rather than buy a 14900K or top-of-the-line Ryzen then kneecap it with a reduced TDP limit.)

Raptor Lake and Alder Lake both had pretty clear base and performance specifications, motherboard manufacturers went well over the performance spec by default(mainly on current limits), and according to Intel themselves "Long term reliability cannot be assured" when going above the maximum values.
The biggest issue is Intel never enforcing those limits, and motherboard manufacturers completely ignoring that Intel themselves don't consider those values reliable, while turning off protections.

Another major issue is the loadline calibration, which Intel has a pretty loose spec, as it's supposed to be calibrated by the manufacturers themselves as it depends on the design of the board. But most motherboards seemingly are just using a set value within spec, which makes some boards be extremely off.

Hopefully this makes Intel and motherboard manufacturers improve their current system. Because clearly there are issues with the way they're currently doing it.


Also 14900K/7950X > other CPUs even when they're power limited, as long as it isn't too low, like 50W and below. For example the 7950X at 88W outperforms the 7900X at stock.

 

[hwertz]

Since we knew that there is a common process in chip production called as "binning" then we should also knew straight away that this is merely an utter example of GREED. In order to ship as many top of the line products they simply relax the binning parameters. Period.

Yup in effect. I don't even know if it's greed but something is going wrong with their binning process, or at least how they calibrate the boost within a bin.

One of the videos I saw, they were having their chip hit like 6.2ghz, unstably (Cinebench was crashing within like 15-60 seconds.) They suspected it could have been stable with another 0.1-0.2V being pumped in there, but the voltage was already high; I suspect that's possible, but in some cases additional voltage will not help and the CPU is just not stable at that speed.

I know these Ryzen CPUs, you'll get examples of "the same model" that vary by several 100 mhz in terms of top boost speed, or they can swap 2 CPUs into the same mobo and the voltage may vary by like 0.05-0.15V. I don't know if they're calibrated at the factory or the AGESA firmware somehow assesses things in the field but whichever, if you just play with power limits (and not actually overclocking it directly), it'll avoid clocking up past the speed where it's stable. It kind of appears that Intel just stuck some values in there without actually ensuring any particular CPU is actually stable at those top speeds.

(Given some comments I've read about people also running into trouble with 6000mhz DDR5 in these setups, I'm wondering if it's possible the CPU core itself is stable at that 6.2ghz but the higher appetite for memory throughput is causing instabilities with the memory interface. I suppose the solution might be the same, they'll just have to slow it's roll; but if that's the case they might also be able to slip in a wait state or something and keep it stable at 6.2. Not that I'd recommend it, in the video I saw that thing was running over 1.8V at that kind of speed which seems very high.)

 

[Angga B]

Yup in effect. I don't even know if it's greed but something is going wrong with their binning process, or at least how they calibrate the boost within a bin.

One of the videos I saw, they were having their chip hit like 6.2ghz, unstably (Cinebench was crashing within like 15-60 seconds.) They suspected it could have been stable with another 0.1-0.2V being pumped in there, but the voltage was already high; I suspect that's possible, but in some cases additional voltage will not help and the CPU is just not stable at that speed.

I know these Ryzen CPUs, you'll get examples of "the same model" that vary by several 100 mhz in terms of top boost speed, or they can swap 2 CPUs into the same mobo and the voltage may vary by like 0.05-0.15V. I don't know if they're calibrated at the factory or the AGESA firmware somehow assesses things in the field but whichever, if you just play with power limits (and not actually overclocking it directly), it'll avoid clocking up past the speed where it's stable. It kind of appears that Intel just stuck some values in there without actually ensuring any particular CPU is actually stable at those top speeds.

(Given some comments I've read about people also running into trouble with 6000mhz DDR5 in these setups, I'm wondering if it's possible the CPU core itself is stable at that 6.2ghz but the higher appetite for memory throughput is causing instabilities with the memory interface. I suppose the solution might be the same, they'll just have to slow it's roll; but if that's the case they might also be able to slip in a wait state or something and keep it stable at 6.2. Not that I'd recommend it, in the video I saw that thing was running over 1.8V at that kind of speed which seems very high.)

Relaxed binning parameters (intel) results in large number of chips got the higher designation than they supposed to.
Stricter binning parameters (amd) ensure variability do not exceed the designation.
IMHO, the quickest solution for Intel had they been too relaxed in binning and have too much stock to sell is to create a new SKU for core i9 14850k or below. Anyone remember core i9 10850k? I do, I used to have one.
And yes, despite your benefit of the doubt the reason is simply GREED. Nothing more, nothing less.

 

[poohbear]

All this x86 CPU and their inane power limits nonsense will become obsolete when ARM CPUs arrive. 25 hours battery life on your laptop while staying within smartphone level power limits? Why yes please! 150wtt PSU enough for an entire Desktop PC? Oh yes!

 

[jamesjpeacockiv]

I'm sitting here on a 13900ks, and since installing it back in October on a Asus Z790 Taichi Carrara, it's definitely falling off. My XTU scores have dropped almost 10% since I first built this system, and nothing internally has changed. I can't even budge the speeds off stock without it crashing immediately after. And power source isn't an issue, as I'm running a 1600 ATXi Corsair. I've been loyal to Intel for over 30 years, but if Intel doesn't start making news for advancements and stability, instead of scandal, bugs, and refreshes, my next build will be an AMD.

 

[maxxcool7421]

Intel did this once already, the P3 1.13GHZ cpu from intel could not run correctly after AMD beat them to 1ghz way back when ..

The cpu should enforce baseline and force users to download a bios unlock to unlock '' unlimited'' power modes, blowing a fuse in the cpu indicating the user agreed to end their warranty

 

[toooooot]

I'm sitting here on a 13900ks, and since installing it back in October on a Asus Z790 Taichi Carrara, it's definitely falling off. My XTU scores have dropped almost 10% since I first built this system, and nothing internally has changed. I can't even budge the speeds off stock without it crashing immediately after. And power source isn't an issue, as I'm running a 1600 ATXi Corsair. I've been loyal to Intel for over 30 years, but if Intel doesn't start making news for advancements and stability, instead of scandal, bugs, and refreshes, my next build will be an AMD.

13700k here. And first disappointment with intel in many years. I should have waited few months and gotten a newly released at the time AMD cpu...

 

[erickmendes]

Here take that car. Breaks not included, drive safely!

 

[user556]

Here take that car. Breaks not included, drive safely!

Brakes, I suspect.

 

[Aranarth]

All this x86 CPU and their inane power limits nonsense will become obsolete when ARM CPUs arrive. 25 hours battery life on your laptop while staying within smartphone level power limits? Why yes please! 150wtt PSU enough for an entire Desktop PC? Oh yes!

That's not how it works.

BTW: all x86 chips in their heart are already risc.

If you want that level of power usage its already available in the x86 world. They are used in laptops and tablets all the time.

 

[Strawman]

Raptor Lake and Alder Lake both had pretty clear base and performance specifications, motherboard manufacturers went well over the performance spec by default(mainly on current limits), and according to Intel themselves "Long term reliability cannot be assured" when going above the maximum values.
The biggest issue is Intel never enforcing those limits, and motherboard manufacturers completely ignoring that Intel themselves don't consider those values reliable, while turning off protections.

Another major issue is the loadline calibration, which Intel has a pretty loose spec, as it's supposed to be calibrated by the manufacturers themselves as it depends on the design of the board. But most motherboards seemingly are just using a set value within spec, which makes some boards be extremely off.

Hopefully this makes Intel and motherboard manufacturers improve their current system. Because clearly there are issues with the way they're currently doing it.


Also 14900K/7950X > other CPUs even when they're power limited, as long as it isn't too low, like 50W and below. For example the 7950X at 88W outperforms the 7900X at stock.

I feel like you are the only one knowing the hell he is talking about. Light load and LLC is the major culprit, mobo manafacturers can't even calibrate their mobos properly and they just use Intel's rec specs. The only mobo I've had that managed to calibrate on auto settings is the z690 unify X.

 

[Gastec]

Great: Intel now has to spend an hour tending to some gamer-troublemakers and ytubers who can't read the instructions for a motherboard or CPU.

Not me! I'm always reading the instructions manuals and all the EULA's for each computer component and piece of software, including video game, before using them. I usually do it in bed before going to sleep, it is such a relaxing lecture!