Are Intel's i9-13900k's and -14900k's Crashing at a Higher Rate? (techradar.com) 66
"Intel's problems with unstable 13th-gen and 14th-gen high-end CPUs appear to run deeper than we thought," writes TechRadar, "and a new YouTube video diving into these gremlins will do little to calm any fears that buyers of Raptor Lake Core i9 processors (and its subsequent refresh) have."
Level1Techs is the YouTuber in question, who has explored several avenues in an effort to make more sense of the crashing issues with these Intel processors that are affecting some PC gamers and making their lives a misery — more so in some cases than others. Data taken from game developer crash logs — from two different games — clearly indicates a high prevalence of crashes with the mentioned more recent Intel Core i9 chips (13900K and 14900K).
In fact, for one particular type of error (decompression, a commonly performed operation in games), there was a total of 1,584 that occurred in the databases Level1Techs sifted through, and an alarming 1,431 of those happened with a 13900K or 14900K. Yes — that's 90% of those decompression errors hitting just two specific CPUs. As for other processors, the third most prevalent was an old Intel Core i7 9750H (Coffee Lake laptop CPU) — which had a grand total of 11 instances. All AMD processors in total had just 4 occurrences of decompression errors in these game databases.
"In case you were thinking that AMD chips might be really underrepresented here, hence that very low figure, well, they're not — 30% of the CPUs in the database were from Team Red..."
"The YouTuber also brings up another point here: namely that data centers are noticing these issues with Core i9s."
More details at Digital Trends... And long-time Slashdot reader UnknowingFool wrote a summary of the video's claims here.
In fact, for one particular type of error (decompression, a commonly performed operation in games), there was a total of 1,584 that occurred in the databases Level1Techs sifted through, and an alarming 1,431 of those happened with a 13900K or 14900K. Yes — that's 90% of those decompression errors hitting just two specific CPUs. As for other processors, the third most prevalent was an old Intel Core i7 9750H (Coffee Lake laptop CPU) — which had a grand total of 11 instances. All AMD processors in total had just 4 occurrences of decompression errors in these game databases.
"In case you were thinking that AMD chips might be really underrepresented here, hence that very low figure, well, they're not — 30% of the CPUs in the database were from Team Red..."
"The YouTuber also brings up another point here: namely that data centers are noticing these issues with Core i9s."
More details at Digital Trends... And long-time Slashdot reader UnknowingFool wrote a summary of the video's claims here.
And probably will apply to the 15900K too (Score:2)
Next refresh looks pretty similar from what little leaks there has been.
Re: (Score:2)
Kind of a redux of use linux problem solved isn't it
Re: (Score:2)
Not quite, switching from Intel to AMD doesn't limit the user in any way. If anything, they get better cost/performance with the current generation.
Re: (Score:2)
Even simpler solution. Don't overclock the fucking CPU into unstable configurations.
You'll note it's on the K series mentioned, the unlocked and overclockable CPUs.
What's the sign of too high of an overclock? The CPU is unstable. Usually tested in GAMES. You push till you crash, then back off a bit...
Is the next news your balls hurt when you smash them with a hammer? And maybe you should have use a rubber hammer instead, because it might hurt a tiny bit less?
Re:And probably will apply to the 15900K too (Score:4, Informative)
Yeah, you've no idea what you're talking about.
This isn't about overclocking. It's crashing even in server configurations with very conservative setups.
https://www.youtube.com/watch?... [youtube.com]
https://www.youtube.com/watch?... [youtube.com]
Re: (Score:3)
A "very conservative" setup isn't going to be running unlocked clock processors designed to be pushed to their limits. You are at the mercy of your motherboard and BIOS / UEFI vendor to ensure your "conservative" setup is within recommended specs at that point. And wasting money as well. If you are running K processors at "very conservative" specs you might as well run the non-K with it's much lower power envelope. You will get better price per watt, have better thermals, and not lose any performance, as ev
Re: (Score:2)
I thought consumer CPUs like the I7,I9 etc are officially not rated for 24/7 use, like in a server?
Not that it hasn't stopped me from using consumer CPUs for 24/7 operation. Even have one of my laptops running 24/7 for almost a week now (with appropriate air flow, etc, of course).
Re: (Score:2)
A "very conservative" setup isn't going to be running unlocked clock processors designed to be pushed to their limits.
Wow you apparently did not even watch the video at all. Level 1 Techs looked at the CPUs used in servers. The keyword is "server" and the CPUs are not "pushed" to their limits in that setting. They are not overclocked. If anything they are underpowered to maximize reliability and stability.
You are at the mercy of your motherboard and BIOS / UEFI vendor to ensure your "conservative" setup is within recommended specs at that point.
Again, you didn't watch the video. The server motherboards have no overlock settings. They are server motherboards.
If you are running K processors at "very conservative" specs you might as well run the non-K with it's much lower power envelope.
Again, you didn't watch the video. That is not true of SERVER settings.
While not impossible for it to be the chips, it's MUCH more likely to be a setup issue.
Again, you did not watch any of th
Re: (Score:2)
If you are paying extra (K series) to overclock the CPU, better make sure the CPU at least have some sort of thermal / safety features that drops the speed if it's getting too close to the edge.
But doesnt the general CPU (AMD and Intel) in recent years have thermal throttling build in?
For me, am using an AMD CPU which is not overclocked. But I did do overclocking many years ago, in my younger days. Now I just get a CPU which suits what I want to do and not hope to overclock it to get to what I want it to do
Re: (Score:2)
> Next refresh looks pretty similar from what little leaks there has been.
Ooof. I had to move my PC into another room due to the heat generated with 14.9k, is it the same with 15.9?
Re: (Score:2)
Re: And probably will apply to the 15900K too (Score:2)
Re: (Score:3)
New core design, new process. Keep in mind that the 12900k/Alder Lake didn't have this problem. Arrow Lake-S (which is what the 15900k will be, if that's what Intel calls it) may not have this difficulty. The worst thing about Arrow Lake-S is that it probably won't be that much faster than a 14900ks/Raptor Lake Refresh. The downside to the 14900ks (and 13900k/ks before it) is that, apparently, it's a deeply flawed CPU.
Yes, Intel deserves a lot of hate for this problem. Yes, you should be spending your
Re: (Score:2)
Raptor Lake-S is degrading even when not "overclocked", never mind that most Z-series boards will unlock boost clocks and power limits for CPUs by default.
Re: The 30% of fails for AMD ... (Score:1)
I remember back when the K6 was a thing we got a bunch of patches for compatibility with it because it behaved differently from the Intel processors and everyone wanted to use the Intel compiler because it had the best output for x86.
These days it doesn't any more, gcc is better, so no more problem there...
Re: (Score:3)
Re: (Score:2)
It's amazing how people forget about that incident or forget that it was proven by analysis of the generated code, complete with a patch that disabled the disabler Intel built in to the start-up code.
Re: (Score:3)
Where do you get 30% from 4 of 1584?
Re: The 30% of fails for AMD ... (Score:2)
Re: (Score:2)
Troll.
i9-13000K Owner (Score:5, Interesting)
I have a dodgy MSI i9-13000K build I got a great deal on. I've done multithreaded video encodes, blender renders, and hours-long Linux compiles, and have had one unexplained crash over the course of about a year.
I did, however, go into the EFI and reduce the maximum CPU power draw from the MSI-set 4000(!) to the Intel recommended ~250W.
Re: i9-13000K Owner (Score:5, Interesting)
It is probably not every processor or we would have noticed way before that. It is likely a production issue. Probably they had bad wafers.
But is could also be something like under high temperature and if you are on a lower quality wafer, this particular instruction fucks up.
I had a few of pentium 3 on which SSE instructions fucked up at high temperature. (sse sqrt if I remember correctly.) But most of the one I had access to worked just fine, even at high temperature.
Re: (Score:3)
I had a few of pentium 3 on which SSE instructions fucked up at high temperature. (sse sqrt if I remember correctly.) But most of the one I had access to worked just fine, even at high temperature.
The extended instructions seem to be an ongoing issue on Intel. AVX512 had all kinds of problems as well, to the point Intel basically punted AVX512 on consumer chips and is doing the hybrid VNNI think instead. Also the TSX instruction set on Haswell, though instead of causing overheating I think it simply didn't work properly.
Re: (Score:3)
The key problem of AVX 512 is that only few problems can actually use vectors of that size. To make the instruction set more useful, they added a bunch of in-vector processing to do various forms of shuffling, data reordering, sparse load/stores kind of things. And these are actually costly in term of complexity of the chip you need to design. But if you don't have them it kinds of becomes hard to use for such large vectors. Without even mentioning the need for 512 bit registers which are quite energy consu
Re: (Score:3)
Not technically "production" but rather a "binning" issue. All wafers are varying degrees of bad, as are the chips on them. They are graded and binned into product lines based on performance and stability at that performance.
Could very well be an incorrect binning.
Re: (Score:2)
It's estimated to be around 25% of affected models.
It's not really clear if heat makes it worse. More cores probably does, simply due to probability.
Intel will try to patch it, but if the patch hits performance people won't be happy with the CPU lottery it creates.
Re: (Score:2)
There is something else going on. It is not bad wafers because errors occur on different CPUs purchased months apart. Also, other products made from those same wafers have had no issues. The issue is also not with temperature as shown by the errors occurring in the data centres where there is appropriate cooling and no overclock.
This appears to be an issue of silicon degradation - a worst case scenario for Intel. So there is a section within the i9 product range that is stressed differently from the
Re: (Score:2)
I read TFA along with a few related ones and it appears that your solution alleviates the problem somewhat, but the rate of failure is still far higher than with older Intel processors or AMD equivalents. What makes things worse is that datacenters are also reporting problems, datacenters do not overclock, in fact this problem is occurring on MoBos which do not even have that option.
A couple of months ago it appeared that the problem was down to overclocking - with Intel having failed to tell the MoBo manu
Re: (Score:2)
I read TFA along with a few related ones and it appears that your solution alleviates the problem somewhat, but the rate of failure is still far higher than with older Intel processors or AMD equivalents. What makes things worse is that datacenters are also reporting problems, datacenters do not overclock, in fact this problem is occurring on MoBos which do not even have that option.
Agreed. If you tell someone not to do something because it will cause problems, or if you tell them, if you're going to do thi
Re: (Score:2)
I had to RMA my first Ryzen 3xxxx, there was a pretty big stink done about them too iirc.
Re: (Score:2)
I helped a client with a gaming system running the Core i9-14900KS (iirc), that was crashing when trying to launch games. I similarly had to set the power limits to Intel's current recommendations to achieve stability. This was before I had heard about the issues with these Raptor Lake chips.
It also wasn't just games. Part of what tipped me off that I was seeing a processor issue was when the NVIDIA driver (a self-extracting 7-Zip archive) would consistently throw a CRC error at a random point during the pr
Re: (Score:2)
if you run memtest, does the issue happens in different places ?
Re: (Score:2)
Didn't particularly test different archives / compression algorithms. The system could definitely extract some other various archives without error, incl
Re: (Score:2)
You tried First Descendant yet? Also, some applications are going into "compatibility mode" because the Raptor Lake-S systems are disabling AVX/AVX2 due to instability.
Elevation (Score:5, Informative)
A computer company I worked for had a problem with our CPUs crashing. Some investigation found the problem much more prevalent at a customer in Flagstaff Arizona, and again at Mexico City.
Cosmic rays were causing problems because we only had error detection and not error detection and correction circuitry in portions of the cache memory design.
Re: (Score:3)
Trace widths are getting small enough . . . (Score:2)
Re: (Score:2)
Re: (Score:2)
Yep, exactly. Design-wise Intel has been behind AMD for quite a while, but their manufacturing edge kept them somewhat competitive as did the crap they did with grossly insecure speculative execution. That is over. Seriously, buy AMD. Intel is way overpriced and unreliable on top.
Re: (Score:1)
I wonder if they're starting to run into unanticipated quantum effects?
No, almost certainly not in the way you are thinking, and definitely not in the way you worded it for others.
Chip fab reached the point allowing for quantum tunneling about 15 years ago.
The FET standard layout libraries have long since been adjusted to both account for preventing tunneling when not desired, and slightly more recently, incorporate designs that intentionally take advantage of quantum tunneling for function.
"Starting to" is the key word prompting the "definitely not" part of my response.
That s
Re: (Score:2)
On top of that, motherboard manufacturers try to make the CPU the fastest by tweaking many of the different settings available. Nobody wants to manufacture the motherboard that runs an expensive CPU slower than what it can run
Pussy Ass E Cores Were The Clue. (Score:1)
Trying to sell silicon that doesn't work anymore. So funny.
I need some stickers that say "AMD Inside".
Article reads like old news (Score:2)
This has been known for a while
Motherboard makers ship BIOSs with extreme defaults, causing processors to run hot. I suspect that they do this to improve benchmark scores
I configured my BIOS settings to what Intel recommended, and all is well
Methinks the problem comes from pushing chips to their limits. Back off a bit, and all is well
In the old days, default settings were conservative and overclockers discovered that they could easily push chips to higher performance
Now, the defaults are at or beyond the li
Re: (Score:3)
Re: (Score:2)
I have a xeon workstation and it came with the same vendor overclocking bullshit enabled as the gaming boards. With modern gen dynamic overclocking largely disappears with 8 and 12 channel systems yet is still around in the lower end 4 channel processors.
Cool story, bro. However in the case of the MBs that Level 1 Techs examined, there are no overclocking settings.
Re: (Score:2)
Intel CPUs have had this problem in one form or another since Coffee Lake/8700k. Alder Lake-S didn't have weird instability and degradation problems despite many mobo OEMs shipping with "defaults" that let the CPU slam up against the 253W PL2 with unlimited tau.
Good old Intel crap CPUs (Score:2, Insightful)
Now that they cannot compete on manufacturing anymore and have been found out about all the crap they did with speculative execution, they apparently try other dirty hacks to fake performance. Seriously, stop buying Intel. You are being ripped off.
Additional Info (Score:3)
A few more details (Score:4, Informative)
According to https://www.youtube.com/watch?... [youtube.com] the processors are not hot, are not overclocked, and processors that will pass a test when new are failing the same test after running half a year.
Re:Statistics? Hello? (Score:4, Informative)
Watch Wendel's video at Level1Techs, and he gives a better representation of the problem. In particular is the data from a company operating Raptor Lake-S CPUs on Intel W680 workstation motherboards as internal game servers. Note that these CPUs do not operate at maximum boost clocks due to the platform (Raptor Lake-S can boost as high as 6 GHz in some circumstances, though W680 seems to keep them at 5.3 GHz and lower).
The company absolutely noticed higher failure rates on Raptor Lake-S CPUs than the previous-gen Alder Lake-S CPUs (12900k/ks). They noticed far fewer failures from AMD's 7950X as well.
Re: (Score:2)
Level1Techs Mentioned (Score:2)
Not just games (Score:2)
In the issues on PrusaSlicer's GitHub page, there have been a couple users that had crashing problems eventually attributed to i9.
These issues pushed me to a 14700k. (Score:2)
I skipped the i9 for issues we are all seeing, and changed my mind about AMD after watching a video from Jazy2Cents, which covered the ongoing memory problems that he was having -- something these i9s are also having now. He also talks about the limitations and issues of the 3D cache outside of gaming:
https://www.youtube.com/watch?... [youtube.com]
I've owned many Intel and AMD CPUs going back decades. My bias it to