How we prevent issues with stability and longevity on the Intel i9-14900k

Published 2024-06-22
It is all about your cooler. Only ask your cooler to do what it can, and nothing more.

Intel is going to release a fix around mid-August 2024 that will prevent high voltages. This will be via a BIOS update from the board makers.

All Comments (18)
  • @xforcepc
    I finally did some research on Aida64. They have so many versions. The version I need, for testing on many machines by one "engineer" is Aida64 Engineer, which I now own. Again, the Engineer version allows one person to put in on many machines for burn-in testing like I do. The "Extreme" version is made for one person to use it on their home PC.
  • @somewaresim
    Seen a few videos showing motherboard manufacturers ignoring CPU recommended power limits. The assumption for most users is that Auto means the CPU vendor recommended settings. Clearly that is not the case.
  • The most recent posts talk about a degradation of I/O portion of the CPU, basicly when Intel designed the Raptor Lake they moved the I/O portion of the CPU to another physical position inside the CPU and that part (called Ring or Uncore) is proving to be uncapable of handling the OUT OF THE BOX voltage settings. The mayority of issues are appearing in i9's and the usual maximum vcore for those parts is around 1.52v ish for the single core advertised 5.8-6.0 Ghz.
  • @ColinDyckes
    A GOOD closed loop, and now a MoRa 420, DOES tame my 14900KS at 320W Extreme profile. Water temperature is 2-3C above ambient.
  • @geesehoward7261
    sure it's possible. You need watercooling. Ive run mine at 370w under cinebench and its been fine. Not for extended times, however. they need to be watercooled and by large custom loops.
  • @phillipzx3754
    An external water chiller for a laser works great. It's a bit spendy but you can keep the CPU nice and ... chilly. :-)
  • @Hyperion1722
    Intel advertised 6 ghz boost. SO does this mean that this is now false advertising??
  • @LiveType
    Not a power/heat problem. What you're doing will still cause issues depending on the motherboard default settings and workloads running. The problem is voltage. Intel is redlining these chips to squeeze as much frequency as they can out of them and they degrade faster than whatever internal models/risk calculations predicted likely due to real world variability being higher than anticipated. These things run 1.5V continuously. If you've ever done any sort of overclocking 1.5V for modern process nodes is a) impossible to cool unless you do direct die w/custom loop or run a chiller for cooling b) will degrade within 1 month of operation typically even 1 week will be enough to show signs of degradation Intel had the ability to "win" against AMD's zen 4 and this was how they did it. Running these chips overclocked as far as they'll go (to hit those boost frequencies) out of the box with a very small amount of wiggle room. That wiggle room is how they compensate for crap motherboards and real world variability and the like. Not having validation of the entire chain with such tight margins is recipe for failure. How intel thought that was a good idea is beyond me. Some chips can of course go even higher and will have zero issues within their lifetimes but those are custom tuned, running direct die, and got lucky. Silicon lottery is a thing. It didn't disappear no matter what any company will tell you. Raptorlake i9 chips running minecraft servers in datacenters clock in a 100% defect rate never going above 65c. Minecraft is the perfect example because it's almost completely single threaded thus won't ever hit high power or temps. But it'll hammer those 6ghz cores with 1.5v+ at load 24/7. That's a recipe for degradation. The solution was to drop the max frequency to 5.5-5.7ghz and all of a sudden failure rates go from 100% to sub 5% (not 0% as they're probably still running degraded silicon/not RMA'd). This lower frequency also results in lower voltages which slows degradation. The power limits you're setting do something similar but only at load. TLDR: Intel pushed these chips too hard from factory to compete against AMD and are now paying the price. They did this before with exactly the same results back in the early 2000s when AMD was also competitive.
  • If you don't want to put a chiller in your system then why is it okay to put a volcano in it? I'm curious as to the reason for getting a 14900K over anything else?
  • @user-jw8er7lg8m
    Seems weird that the processor can't control how much power it consumes and needs hard limits from the motherboard.
  • @shodan6401
    No, no, no - just no. This issue is very, very simple. To get top scores in some reviewers benchmarks, like Cinebench, Intel allows a single core (or two) to boost to 5.8 or 6.0GHz. To reach that freq. the core gets 1.55 or even 1.6v - which causes damage. This is NOT a board problem. Intel endorsed these power settings because without them, Intel can't claim 5.8GHz. So blaming the board is BS. Intel said on record that it is only considered out of spec. if you change the clock multiplier. Intel is the source and cause of the issue. The fix is easy: SYNCH all your cores. Don't allow one or two cores to boost by themselves. I've seen a PC in an idle state suddenly BSOD because a background task started. Then a core tries to boost to 6.0GHz, gets 1.6v and now you have a degraded CPU. Synch the cores. Might have to lower your multiplier depending on the state of your chip, but after that you're safe. Problem solved.
  • @pvdgucht
    X-Plane 12.1 still uses only 1 core at 90% on a 14700K and the other cores are chilling. So 🤷🏼… personally I have a small air cooler on my 14700K and its just fine if u don’t use 100% of the CPU like in any game or flight sim. So unless X-Plane is gonnan be improved to really take advantage of every core well untill then I don’t even need to buy a water cooler aio.
  • trial version isn't that suicide for commercial use? license allow for that?
  • @CAL1MBO
    I wouldn't use the Heaven benchmark you have on the desktop. It's barely acceptable to benchmark a modern 1080 system, it's completely useless for 1440+ or testing stability for any overclock. I recommend 3d mark.
  • Just to mention, an Intel engineer stated point blank that you are allowed to ride 100c on these CPUs, therefore not leaving any performance on the table.
  • @Exodus.1337
    Damn this is sad if you listen to this guy.. not the fix lmaoo
  • @arionflorin4257
    Yes, get a K cpu and undervolt it or limit its power, might have been better to get a non K cpu or a 14700 without K. amd and intel both are at fault, for the last 2 generations its more power and more heat.