Intel 14th/13th gen, latency/stability issues?

[anoise]

But I have to say, this thread director and/or Win11, still puts Bitwig on the E-cores if the window/application is not in focus. Play a track in bitwig with the DSP perf graph open, then switch to another app like a browser and use that for a few seconds, you will see in bitwig the performance tanks and audio crackles, because the thread director or win11 decided to put Bitwig on the e-cores.

Anyone seen this behavior?

I don't know if this is a Bitwig issue or not. I need to ask them if they are aware of this. But I read others have the same issue with different software...

https://www.reddit.com/r/WindowsHelp/co ... ance_when/

[Pictus]

Finally official numbers for 14700k, took them a whole to write those.

Meanwhile I had my first bluescreen. W10 gets stuck on welcome screen out of the blue, what an abomination of a system.
And if I run a really heavy Kontakt instrument I get crackles. It feels less and less an upgrade.

You have the ASUS HERO, so do not install ASUS armory crate and in the BIOS try:

- SB Clock Spread Spectrum = Disabled
- Power Supply Idle Control = Typical Current Idle
- VDDCR CPU Switching Frequency = 350 (If the default is higher, keep the default)
- VDDCR CPU Power Phase Control = Extreme
- VDDCR SOC Switching Frequency = 350 (If the default is higher, keep the default)
- VDDCR SOC Power Phase Control = Extreme
When Power Phase Control = Extreme, all VRMs will stay active, they will not
"sleep" even in idle, this generates more VRM heat, but better electric signal.
Because the Hero motherboard has an excellent VRM you do not need to use
EXTREME for the CPU, but should use for the SOC.

- Do not undervolt the CPU.
- Do ALL my tweaks.
- Use Windows 11.
- Try to reseat the CPU.
- Verify if the CPU socket have bent pins.
- Do not overclock the RAM.
- Do not overclock the CPU.

The RAM speed maybe is too much for the memory controller.
The Intel memory controller sometimes is a bitch and hard to detect errors.


Test the RAM/memory controller, in your case leave it running 6 hours or more
***ONLY*** the y-cruncher tests 17/18/19/20.
The other tests will "fry" the CPU! You should not do the other tests for hours and
hours, but can do for minutes.

-Type 2 + enter
-Type 8 + enter
-Type 17 + enter
-Type 18 + enter
-Type 19 + enter
-Type 20 + enter
-Type 0 + enter

The CPU may be defective.

[Pictus]

My company has been building DAWs professionally for over 30 years.
There's nothing wrong or sub-par with the i9 13900k or 14900k.
Yes, they draw a lot of power.
Yes, they require top-tier 480mm AIO cooling.

I guess you will build more AM5 now.
Current AM5 users can upgrade by only replacing the CPU.
AMD Ryzen 9 9950X CPU Outperforms Intel Core i9-14900K In Blender At Just 120W
https://wccftech.com/amd-ryzen-9-9950x- ... r-at-120w/

[Pictus]

Intel's 13900k's and 14900k's are crashing at an alarming rate?

[PAK]

Intel's 13900k's and 14900k's are crashing at an alarming rate

Yeah. Not good. Chips appear to start out ok but "degrade" over time. It's also looking like the earlier "motherboards running chips out of spec" problem might've been a bit of a red herring, and actually alluded to this wider problem (The failures are also being seen on server boards which have never ran at the higher voltages.. )

Wendell (of L1) estimates that, conservatively, Intel might have to end up replacing at least 25% of the CPU's - and even that assumes they provide updates which can mitigate the problem for others. Apparently Gamers Nexus also think they have an idea of what the issue is, so I guess we can expect to see some more videos on this soon.

At this point people shouldn't build systems based on the 13900 / 14900. If you do expect instability which increases over time, and which may be made worse by other factors like higher DDR speeds. I assume Intel will do the right thing, and replace any chips which can be shown to have problems. The longer you have run the chip, the more likely that is to be true.

[Zombie Queen]

I did a bit of gaming. And yep I have crashes. Early to draw conclusions, as it's just one game, but it's like random crash every 2 hours. Initially I blamed overheating GPU, but nope. Eventually I did what the guy in second video said they did (so it resonated for me), reduced memory clock. Again, too early to say, but for a first time it didn't crash.

[Pictus]

I did a bit of gaming. And yep I have crashes.(...)

RMA the CPU.

[Zombie Queen]

RMA the CPU.

Nah. Too much trouble, it would have to get seriously bad. One game crashing is not nearly close to the threshold of RMA, especially considering swapping CPU may not help. Changed memory clock from 6400 to 6200, changed XMP profile from II to I, no crashes since.

Intel is in breaking point situation IMO. Either they will do something bold, or they go down. If I was to bet, I would say they will do nothing and eventually they will fall down to third position in CPU manufacturing. Sadly.

In another video there was a suggestion it's all due to moving I/O hub to separate section of CPU. So possibly a fundamental design flaw.

[Pictus]

[Jim Roseberry]

Hyperbole to get folks to click on YouTube videos.
The most likely cause of instability is an inexperienced builder not knowing how to configure the BIOS (including RAM speed).
ie: If you build a 14900k based machine with DDR5/6000... set XMP and think you're done... you're going to have massive stability problems.

CPU failure rates are up (Post Covid vs the previous 20 years)... but nowhere near 100% for the 13900k and 14900k.
I've had to replace three 13900k CPUs... out of scores.
I have not replaced a single 14900k.

[Scotty]

You are generally correct but this is a CPU issue and not an casual builder making simple novice mistakes issue. Follow the topic more closely and watch the headlines. This isn’t a figment of someone’s imagination.

Hyperbole to get folks to click on YouTube videos.
The most likely cause of instability is an inexperienced builder not knowing how to configure the BIOS (including RAM speed).
ie: If you build a 14900k based machine with DDR5/6000... set XMP and think you're done... you're going to have massive stability problems.

CPU failure rates are up (Post Covid vs the previous 20 years)... but nowhere near 100% for the 13900k and 14900k.
I've had to replace three 13900k CPUs... out of scores.
I have not replaced a single 14900k.

[Jim Roseberry]

You are generally correct but this is a CPU issue and not an casual builder making simple novice mistakes issue. Follow the topic more closely and watch the headlines. This isn’t a figment of someone’s imagination.

I hear what you're saying...
FWIW, I build a fair amount of machines.
Not one, two, or three a year...
To claim that the failure rate for 13900k and 14900k CPUs is anywhere close to 100% is exaggeration (hyperbole aka click-bait).
As I said, out of scores of 13900k and 14900k based machines, I've replaced three bad 13900k CPUs.

Granted, that number should be closer to zero.
I've seen as many bad CPUs post Covid as I had the previous 25 years (pre Covid).
QC is definitely worse.
But, c'mon... realistic percentage is far lower.

[Zombie Queen]

The video which made a lot of fuss, was one where there was 'unofficial' data from a game server center, claiming 50% of their 14900k based systems experience crashes. So this is a server controlled environment, no overclocking, W680 mobos. Also it looks like problems manifest in specific tasks - like gaming: user end and server end. So it may also be your client profile bias (not gaming).
I didn't see any problems until I launched a game. DAW, video editing, AI training - 0 problems. Basically I had to slightly reduce memory clock, and I seen no gaming problems since. See if it stays that way. Somehow I am sure we will hear more and more about it.

Jim, did you upgrade BIOS to one released couple days ago, patching CPU microcode? I remember you mentioned running Z790 Hero...

[Zombie Queen]

100% failure rate source is this article in PC Gamer.

[anoise]

The degradation, if true, is concerning. My 14700K seems stable (I did not test every possible scenario), had no issues with undervolting. Let's hope it does not degrade through the years/months.

Question to the experts here, like Mr. Jim Roseberry, are the following settings and voltages healthy? I think so, but a second opinion would be great.

My settings on ASUS Z790-P:
PL1=PL2=253W
ICCMAX: 280A (Noctua NH-D15 air cooler)
SVID: Typical, LLC: 3 default (AC/DC=0.400/1.100)
Voltage Offset -0.040v
CEP Off
DDR5 XMP 6000 CL30 @ 1.4v

With these settings the ~maximum voltages are:
Idle: VIDs: 1.367v, Vcore: 1.350v
Full load (CB23): VIDs: 1.127v, Vcore: 1.172v = ~85c, CB23 score ~33716 (sometimes it hits 34)