Gigabyte, MSI are tackling Core i9 crashing issues with BIOS updates and user guides

After majority of motherboard manufacturers fix the problem I think it is worth to retest 14900k/14700k performance.

so, I have a 14900K and I have been worried about these crashing issues. And, like most, I've heard about it being an issue with the motherboards overpowering the CPU. The thing is, even with everything at stock setting, and letting the motherboard do it's own thing, it's still not providing as much power as what it first shows. Especially when you try to run it - it will instantly thermal throttle, even with a custom loop cooling setup. There are a number of videos and articles about people "overclocking" their CPU to 6.2GHz all core or whatever, but in reality, it's never able to run at that, it will always thermal throttle and downclock. I left my clocks as is, and put a 1.35V override on the CPU core, LLC (MSI board) of 6. It doesn't thermal throttle when maxed running Cinebench, and I actually score higher than everyone that is "overclocking".
Their systems may boost very briefly to 6.2GHz or whatever, but it would be way too short of a time to make a difference in actual performance over being able to keep stock permanently and not downclock the rest of the time.

dalek1234 said:

Don't forgot to add that user are now reporting that this "update to resolve crashes" comes with a ~9% performance penalty (including in games)

It could be argued however, that the user should not have had the extra 9% performance. Yes I know that opinion is not popular, but on the other hand, whether intel was involved or not, the proper way seems like for the future would be that manufacturers put things at the Intel stock settings and that is the baseline for performance.

Then they can still do performance profiles etc as an option and have a guide with the motherboard telling people there's this performance profile, here's how you enable it, but this profile is not officially supported and here are the risks of it and put it into the hands of the user if they wish to tinker with or not. Not force that out of the box because you are trying to beat another product to sell more by going outside the specs of the product.

Kind of like if I go buy a Honda Civic and the dealer installs an aftermarket turbocharger that blows up the engine 6 months later and I wasn't aware that part had been installed.

So I looked further into these proposed fixes by the motherboard makers and they are definitely not for everyone.

From what I've heard some CPUs are unstable under some shader compiling of some games where they use a ton of power and crash. And giving the CPUs so much extra volts, in combination with a standard power limit, which pretty much guarantees they will power throttle to lower top clocks, fixes the instability.

It sounds like the affected CPUs are unstable at high clocks under high power use scenarios with other settings being stock. Whether these are 1-2 p-core clocks or 8 p-core clocks I don't know. I'm guessing the higher 1-2 p-core clocks since we haven't heard of chronic crashes with things like Cinebench, CPU-Z and other typical all core benchmarks that run at the 8 p-core clocks when not thermally constrained. But since I don't have this issue I don't really know.

There are a few ways to help the stability of higher clocks that are less brute force than going with Intel's fail safe voltages with stock power limits. They all involve increasing the volts over what is unstable and/or decreasing the unstable clocks. If a lighter all core loads like the CPU-Z benchmark are stable and Cinebench is borderline with these games being unstable then vdroop may be the culprit and the CPU may be drawing enough power on these problematic loads that the voltage droops below stable levels. These i9s are drawing more power than consumer CPUs ever have and maybe the motherboards or power bits on the CPUs aren't up to the task on the stock settings of some CPUs. Some options to address this are:

1. You could try increasing the per point voltage offsets at the higher clocks.
2. Reduce all volts and raise the load line calibration preset so that you get about the same volts per stable all core load, but more under extreme loading.
3. Cap the power to stock.
4. Reduce the 2 core boost.
5. Have your clocks dependent on temp, like by reducing them and then bringing them back to stock with a Thermal Velocity Boost overclock that will reign them in when they get to either a stock 70c or whatever temp you set.
6. If the culprit load is using AVX, maybe an AVX offset is in order.
7. Turn off that HT that is worthless for everything but all core benches when you have 16 e-cores.
8. Some combination of the above.

The reason why I'm suggesting this is just how much volts Intel's fail safe is. And if the CPU is stable in just about everything it is probably very close to being stable in everything.

I tested it with an Asus Prime Z690P, 13900kf, bios version 3202, new enough that APO is working.
I reset the bios to Asus default optimized settings and for one set of tests I went through all SVID options and with the other 4 load line calibration presets that spanned from the weakest vdroop compensation to the strongest. I used CPU-Z Stress CPU to load my CPU since it is fast and easy to check loaded volts, and I cycled through CPU clocks using Windows power plans that kept the e-core clocks requesting voltages just below that of the p-cores to maintain a proportional all core load.

So here's the ugly:

The top voltage SVID and LLC settings were too rich for my new awesome $51 cooler to handle: https://www.amazon.com/Thermalright-240Black-Specifications-Bearings-Suitable/dp/B0BLJHRVF5/ref=sr_1_1?crid=3FAORYOUIJPBV&dib=eyJ2IjoiMSJ9.Nt6_GIiId4IW39Pjp5lLshCEMGs3wa1bDQmm-gAq2aMnm8QMsH2j_jOnPqAeiJIZ4g2raKnd7H4yLsAi2XGfvK2Mhg8jM4Kcxr9xXYwJDae8HF2dj9CDpwoAAPKr4fxX9IZ2mk_Kwd51BQnMLAXiJbKQUKirZEFP674_53KM1dqpwAQEfO43jrqRPh8lSdn8zGFuJf46DGI-LNg0Yht4JrjaKzWTpeiO6XMQYlU1b7M.Imq7w-tCZFhzMBlskVh_7xlr1fkyba4FBu-bUxU2i6g&dib_tag=se&keywords=thermalright+cooler+aio&qid=1714177430&sprefix=thermalright+cooler,aps,104&sr=8-1&th=1Even running CPU-Z stress. Cinebench probably uses 20% more power. If your CPU is terrible you may need to dump in that much power. But if you don't have to I would avoid it.

I personally use Best-Case Scenario with LLC 6 that increases my volts to halfway between Best Case and Typical at LLC 3 (Asus default on my Prime P) at 5.5 GHz. And I added 20 mv to my e-cores L2.

I really just wanted to point out how much power these settings could waste if you applied them haphazardly. I don't think it would hurt your CPU (I saw 1.62v with Intel fail safe + LLC 6 messing around with this and everything seems fine), but having such a hot CPU would detract from your quality of computing.

Edit: Also your voltages may vary depending on your CPU, motherboard and maybe power supply.

This entire thing is a mess caused by Intel letting motherboard manufacturers do whatever they want for over a decade now. Intel specifications regarding power are largely recommendations based and in the case of loadline settings there's a min/max, but it's up to the manufacturers where they land. There's a good chance that they're being too aggressive with their tuning which isn't allowing the CPUs to get enough voltage under these high clock load circumstances.

I feel like the best solution here is for Intel to step in and force some of these settings to be changed under defaults. Given that none of this is damaging CPUs I don't think they need to enforce hard limits, but the default settings on these boards are clearly insufficient for some of the CPUs which makes everyone look bad.