News Intel offers new guidance on 13th and 14th Gen CPU instability — but no definitive fix yet

[Admin]

Intel continues to investigate what's causing instability problems with its 13th and 14th Generation CPUs, but is rolling out a fix for one contributing bug.

Intel offers new guidance on 13th and 14th Gen CPU instability — but no definitive fix yet : Read more

 

[DS426]

Not to put him on the spot but didn't Jay from Jayz2Cents say he switched from AMD to Intel due to stability issues and Intel "just working"? Anybody appreciate the humor/irony in that besides me?

 

[HaninTH]

Not to put him on the spot but didn't Jay from Jayz2Cents say he switched from AMD to Intel due to stability issues and Intel "just working"? Anybody appreciate the humor/irony in that besides me?

All companies will have their ups and downs. Believing otherwise is short-sighted/fool hearty.

How will Intel bounce back from these issues is my concern. Surely, they will, but will these issues require entirely new silicon?

Will AMD try to capitalize on Intel's travails and make further gains to their overall market share?

How is the overall PC/PC Equipment market doing and is there even any space to make any kind of movement or is the lull enough that Intel will be able to recover before anyone (mainstreamers) notices?

 

[DS426]

All companies will have their ups and downs. Believing otherwise is short-sighted/fool hearty.

...

Absolutely.

Will AMD capitalize on it? Well, they usually fumble the ball when Intel hands them a marketing victory, so I doubt it, lol.

I imagine Intel can squash this just with tuning and soft updates (microcode/firmware), granted they may indeed find at some point a hardware flaw that really let the cat out of the bag on this issue.

Lunar Lake and Arrow Lake will have huge efficiency improvements such that I don't see Intel suffering from the nature of this issue for those chips... but stability issues could always creep up from another angle.

 

[bit_user]

IMO, the most disheartening aspect of all this is that they didn't discover it until well after Gen 14 launched! That's a bad indicator of their internal QA practices and represents a missed opportunity for them to fix it in Gen 14.

 

[bit_user]

How will Intel bounce back from these issues is my concern. Surely, they will, but will these issues require entirely new silicon?

It's much too late for that. They will "fix" gen 13 & 14 processors by mandating safe BIOS settings to board makers. Intel won't bother doing a respin of Raptor Lake, right in the lead up to the launch of Arrow Lake.

Will AMD try to capitalize on Intel's travails and make further gains to their overall market share?

You mean, will they do some ad campaign? Because, it seems like they do benefit by default.

How is the overall PC/PC Equipment market doing and is there even any space to make any kind of movement or is the lull enough that Intel will be able to recover before anyone (mainstreamers) notices?

Seems like Intel could see a dip in their i7/i9 sales until Arrow Lake launches. Fortunately for them, they got through the launch of Gen 14 and holiday season, before this came to light.

I imagine Intel can squash this just with tuning and soft updates (microcode/firmware), granted they may indeed find at some point a hardware flaw that really let the cat out of the bag on this issue.

I think a fix that doesn't compromise performance will not be possible.

... Arrow Lake will have huge efficiency improvements such that I don't see Intel suffering from the nature of this issue for those chips...

No, I'm pretty sure we know PL2 for Arrow Lake will continue to be at 253 W for an 8P + 16E configuration.

 

[thestryker]

IMO, the most disheartening aspect of all this is that they didn't discover it until well after Gen 14 launched! That's a bad indicator of their internal QA practices and represents a missed opportunity for them to fix it in Gen 14.

I honestly don't think they test retail boards. This is mostly a sign that they really need to and also probably need to have more strict requirements for board partner settings.

Not to put him on the spot but didn't Jay from Jayz2Cents say he switched from AMD to Intel due to stability issues and Intel "just working"? Anybody appreciate the humor/irony in that besides me?

Except that he's not reported any issues with his system (which he certainly wouldn't be shy about), and none of this erases all the problems he had on AMD.

 

[jp7189]

All companies will have their ups and downs. Believing otherwise is short-sighted/fool hearty.

How will Intel bounce back from these issues is my concern. Surely, they will, but will these issues require entirely new silicon?

Will AMD try to capitalize on Intel's travails and make further gains to their overall market share?

How is the overall PC/PC Equipment market doing and is there even any space to make any kind of movement or is the lull enough that Intel will be able to recover before anyone (mainstreamers) notices?

Honest question, did this affect anyone in the mainstream? I doubt Dell et.al. would have ever entertained enabling TVB or changing power limits off the baseline.

In my mind, if you buying components to assemble yourself, your already in the enthusiast camp, and (presumably) have the knowledge to modify setting to avoid this problem.

That's not to take away from the disappointment we feel that Intel gave us hope that their procs would perform better while maintaining their frankly well-deserved reputation for stability.

 

[CmdrShepard]

I think a fix that doesn't compromise performance will not be possible.

I disagree with your assesment here because this isn't a bug that affects code execution directly like previous side channel vulnerabilities whose patching led to direct loss of computational performance.

If there's any loss of performance it will come at the expense of boosting past the safe thermal limit (while at 90+ degrees). That's already a part of the curve where going further yields diminished returns at the expense of excessive heat and power consumption (think 3% performance with 50% more power or something to that tune).

 

[jp7189]

In my mind, Intel still moved the bar forward from the 12900k, but to be sure I personally find it an enjoyable hobby to spend hours overclocking stuff.

I had a 12900k that was constantly downclocking itself due to bumping the temp limit despite the water itself never exceeding 35C. I spent a lot of time tweaking, reinforcing the <bendable> socket, swapping water blocks to one with a more convex profile. I dropped a 13900k in to the exact same setup, and it never had temp issues and had no problem reaching and maintaining clocks the 12900k could never. Despite that, I've never noted crashes of any sort, but as I alluded to above I tend to set everything to manual, if for no other reason than to play with each thing.

 

[wingfinger]

Are the non K processors ok and/or it is also related to the Z chipset?

 

[bit_user]

I honestly don't think they test retail boards.

If not, then they must've stopped in just the past couple years:

https://www.tomshardware.com/features/inside-intels-secret-overclocking-lab/5
​

This is mostly a sign that they really need to and also probably need to have more strict requirements for board partner settings.

Maybe, but I'm not convinced there's no aspect of their process node or cell libraries they didn't fully understand. Perhaps a comprehensive fix would've involved some tweaks there.

 

[bit_user]

If there's any loss of performance it will come at the expense of boosting past the safe thermal limit (while at 90+ degrees). That's already a part of the curve where going further yields diminished returns at the expense of excessive heat and power consumption (think 3% performance with 50% more power or something to that tune).

The curve isn't as flat as you think. Raptor Lake has long legs. It needs a lot of power to maximize its performance potential.

​

​

 

[thestryker]

If not, then they must've stopped in just the past couple years:

OC lab != QA validation lab

They test the functionality and limits but that doesn't mean they do long term performance/stress testing using partner/retail boards.

Maybe, but I'm not convinced there's no aspect of their process node or cell libraries they didn't fully understand. Perhaps a comprehensive fix would've involved some tweaks there.

While I don't think we'll ever know for sure depending on whatever the fix may be that may shine a light on what happened.

 

[bit_user]

OC lab != QA validation lab

They test the functionality and limits but that doesn't mean they do long term performance/stress testing using partner/retail boards.

But it means they have retail boards in their labs and surely know what kinds of things those boards are doing. So, they know the envelope they could test, if they wanted to ensure stable operation in the field.

 

[thestryker]

But it means they have retail boards in their labs and surely know what kinds of things those boards are doing. So, they know the envelope they could test, if they wanted to ensure stable operation in the field.

Yes, but it's obvious they didn't test them as part of their long term suite otherwise they would have caught it before 14th Gen launch (whether or not they fixed it).

 

[helper800]

Yes, but it's obvious they didn't test them as part of their long term suite otherwise they would have caught it before 14th Gen launch (whether or not they fixed it).

Although likely true, everyone misses bugs even when they do test for them extensively. Point being, no matter how much testing is done, bugs will slip past the testers. There is no such thing as perfect code/hardware.

 

[cyrusfox]

Truly disappointing that root cause and permanent fix has thus eluded being found. It will be instructive and beneficial to current and likely next generation processors to have this sorted.

On my 13900 non k, I experience zero issues, personally just waiting for Arrowlake top sku. No Guarantee this won't resurface there, especially as exact root cause is still TBD.

 

[helper800]

Truly disappointing that root cause and permanent fix has thus eluded being found. It will be instructive and beneficial to current and likely next generation processors to have this sorted.

On my 13900 non k, I experience zero issues, personally just waiting for Arrowlake top sku. No Guarantee this won't resurface there, especially as exact root cause is still TBD.

Maybe wait a few months, I'm sure a 13900 can tide you over for the sake of possibly skipping such issues.

 

[bit_user]

Although likely true, everyone misses bugs even when they do test for them extensively. Point being, no matter how much testing is done, bugs will slip past the testers. There is no such thing as perfect code/hardware.

Bugs are one thing, but this is a problem that happens on every major enthusiast board (as far as I've heard), meaning it's not a special case. Seems like a lack of stress testing + detailed analysis.

For someone like Intel, they shouldn't need to wait until software failures start happening, like us mere mortals. Intel should be analyzing several parameters of silicon aging and checking whether it aligns with their models or predicts premature failures. That's how you'd be able to know about these problems maybe not before launch, but certainly before Gen 14 was locked & loaded.

 

[bit_user]

Truly disappointing that root cause and permanent fix has thus eluded being found. It will be instructive and beneficial to current and likely next generation processors to have this sorted.

Perhaps, in general. However, these would seem to be on quite different process nodes, so I expect the main takeaways will be more related to testing and keeping their board partners on a tighter leash.

On my 13900 non k, I experience zero issues, personally just waiting for Arrowlake top sku.

OMG, you guys. I just upgraded from a Sandybridge i7 to an i5-12600. Whatever do you do that needs so much performance?

At work, I have an i9-12900. It's great for compiling our huge codebase and lots of other packages we use. We upgrade desktops about every 5 years (or about twice as often as I do at home).

 

[cyrusfox]

OMG, you guys. I just upgraded from a Sandybridge i7 to an i5-12600. Whatever do you do that needs so much performance?

At work, I have an i9-12900. It's great for compiling our huge codebase and lots of other packages we use. We upgrade desktops about every 5 years (or about twice as often as I do at home).

Do I need a top sku/best performance? No I would be fine with a 12400-class cpu. Only occasionally do I force a full 24 thread workload, and it is always performs slower than I would expect...(Still leagues better than the quad and 8-core 9900k of yore). I can live with garbage performance(its what I am use to outside).

Part of it is an excuse to play with the latest CPU, the other part is compensating for how slow computing is I do everywhere else (Thank you corporate software that logs and analyzes every key stroke and every action).

CPU's are cheap compared to GPU's these days, might as well upgrade the more affordable component I am also waiting for P5800x to get to my impulse purchase price range, I think I likely have 2-3 years to go before those come down, if they ever do being the last of its species... Its been about 10 years since I did a custom full loop system, maybe time to do it again.

 

[bit_user]

occasionally do I force a full 24 thread workload, and it is always performs slower than I would expect...

What I do is increase the power limits, slightly. For my workloads, I never see the machine reach its stock 202 W PL2. So, I just focus on PL1 - which I boost to 80 W. I could go to 85 W, but much higher will overwhelm the machine's cooling. Luckily, it's in an air-conditioned machine room, so I don't have to make any noise tradeoffs and that little blower it has does indeed have quite a set of lungs (the machine is a Dell compact desktop). One thing that's interesting about raising PL1 is that it actually extends the PL2 boost period, due to the way their power management algorithm works.

Also, how do you have only 24 threads on an i9-13900? Disabled hyperthreading?

Part of it is an excuse to play with the latest CPU, the other part is compensating for how slow computing is I do everywhere else (Thank you corporate software that logs and analyzes every key stroke and every action).

Oh, that. Yeah, my corporate laptop runs Windows and it does bog down sometimes. Luckily, I don't need to use it for development.

I am also waiting for P5800x to get to my impulse purchase price range,

I bought one (400 GB), but still haven't used it. I'm just not sure I would ever be able to feel a performance difference. I think I might just stick with the P5520 I snapped up, during the SSD market crash.

I thought the P5800X just might increase in value, like the Radeon VII I sold to fund its purchase. Instead, they're on ebay for half what I paid. I considered that possibility, too. Then again the supply of new ones on ebay should eventually dwindle and then we might start to see prices go back up.

 

[TheHerald]

I think a fix that doesn't compromise performance will not be possible..

I think the fix will increase efficiency by a lot. Win win situation, no?

 

[TheHerald]

The curve isn't as flat as you think. Raptor Lake has long legs. It needs a lot of power to maximize its performance potential.

​

​

​

​

The fake graph AGAIN? I mean come on now man.....how many times do we need to see this anandtech megafake graph, when the reviewer himself said it's problematic?

You are not helping amd by doing this, imagine someone actually buying the 7950x - setting it to 65w and expecting to see 31k.