Discussing thermal throttling and undervolt concerns for the i9 13900K paired with Arctic Liquid Freezer II 420
Discussing thermal throttling and undervolt concerns for the i9 13900K paired with Arctic Liquid Freezer II 420
I operate an i9 13900k paired with an Arctic Liquid Freezer II 420 on an MSI Z690 Carbon WiFi. With default BIOS settings (PL1/PL2 unlocked), I achieve approximately 38200 on Cinebench R23 MC. The power consumption stays near 300W. Almost instantly, the p-cores hit temperatures between 97-100°C, with p-core 0 and 1 consistently around 88-92°C. Consequently, these cores slow down to about 5.2 to 5.3 GHz. On Prime95 SmallFFT, power usage rises to around 330W, but the p-cores drop to 4.9-5.0GHz. When I set PL1/PL2 to 253W, the p-cores maintain roughly 5.0-5.1 GHz and temperatures drop to about 85°C during Cinebench R23 MC, yielding a score near 36500.
My first concern is whether these readings match typical expectations. My second thought is about undervolting discussions found on Reddit—some users mentioned -0.1V, others -0.05V with minimal impact. For me, anything below -0.015V (after enabling Offset Core Voltage Mode and adjusting to -0.015V) leads to instability. This includes full system restarts during Cinebench or crashes after a few seconds of use. The -0.015V setting slightly improves the score by about 1% but doesn’t change temperatures or throttling much.
It seems unlikely I’m uniquely positioned for such low voltages; are there MSI BIOS defaults I should adjust? Ambient temperatures sit between 24-25°C. My hardware includes an Arctic Liquid Freezer II 420 Fractal Design Define 7 XL, MSI MPG Z690 Carbon WiFi BIOS 2x32GB G.SKILL Trident Z5 RGB DDR5 6400MHz (stock at 4800MHz due to XMP profile instability), Corsair HX1200. I’ve opened the case completely for testing, but cooling still causes throttling during Cinebench R23.
This will receive significantly more fresh air compared to other methods.
They're a little lower than what I'd expect on average, but different boards and CPUs have different stock voltage behavior and can dramatically change power draw and temps so this isn't impossible. I would want to check the thermal paste application, maybe use a bit more just in case, but it wouldn't surprise me if this was just how it ran. In my experience (3 different 13th gen chips, 2 13600Ks and a 13700K), the range for undervolts is anywhere from -0.150V (god-tier 13600K) to 0.050V (middle of the pack 13700K) all on a Z690 Unify-X with a pretty low stock voltage behavior. That's by no means an exhaustive list, but it would surprise me that -0.015V wouldn't be stable at least in Cinebench (Cinebench is relatively easy to run, the same voltage I need for fully stable 5.6GHz on that 13700K can do 5.8GHz in Cinebench all day). It doesn't sound like you're doing anything that wrong, that should be all you have to do to get the chip stable on an MSI board, though you might be better off messing with just setting up a static voltage. The static voltage method will be harder to setup than just a negative offset, though in my experience it can get a little better results albeit at the cost of some idle power draw (not a ton though). The way to do it is as such: Go into the BIOS and head to the DigitALL power settings in the BIOS. Set the LLC to level 7 (on almost every MSI board this is optimal for voltage regulation). Change the VCore mode to manual and set it to something like 1.3V. This value will be higher than what it actually runs because of the way LLC works and this value is just a baseline. Go into Windows and run Cinebench, and check for thermal throttling. If it thermal throttles, lower the voltage until it stops. If it doesn't throttle, make sure the score is as expected (~40k points in R23) and start lowering the voltage until you start crashing. Once you're at the limits in Cinebench, move onto an actual stress test like Linpack or Prime95. Run them, check for thermal throttling, then start raising the voltage till it is fully stable. That's the method I've done it pretty much every time, it's a little more consistent than the negative offset method (albeit not a ton). Plus I'm from the old school OC days, and that's more or less the method for a traditional overclock, so take that for what you will.
Thank you for sharing the details. I've reviewed your situation and noticed the HWINFO64 shows a core voltage of 1.300V with a small fluctuation. This suggests the system is running at the expected voltage but may be experiencing instability during high load. The reset to default settings (F6 - Load Optimized Defaults) and adjusting LLC to 7 with Core Voltage set to "1.3V" seems like a reasonable approach. Since Cinebench R23 crashes right after starting an MC run, it could be related to thermal throttling or driver issues. Consider checking the cooler mount stability and ensuring proper ventilation. If the problem persists, further diagnostics with a different power supply or cooling solution might help.
I won't be able to make the pump the lowest point of the loop. I checked with side panels off and saw only a small difference. The air moving into the radiator isn't very warm, and the air leaving it (and the radiator) isn't either. However, I heard that this might be typical for a 420mm AIO.
This setup isn’t ideal, but it should still function. Sufficient paste should be present for basic use. It could help to invest in a contact frame, which would improve temperature management. Possible causes include focusing on the VID instead of VCore or incorrect LLC configurations on the board. I recall someone else experiencing similar problems, where the LLC settings appear reversed compared to others. The VCore value should decrease to around 1.1–1.2V under load, not remain at 1.3V. With level 7, I wouldn’t expect idle voltages to exceed 1.25V. Consider adjusting the LLC settings—try level 2 or 3 instead of level 7—to see if voltage drop improves. If that doesn’t work, you may need to raise the voltage to approximately 1.35 or 1.4V for testing. Regardless, you’ll likely need to boost the voltage from 1.3V if Cinebench performance isn’t met.
The issue stems from the VCore setting and thermal behavior. It seems the high idle voltage and aggressive throttling are normal under load, but it’s pushing performance limits. You might benefit from adjusting the profile to prioritize stability over raw speed, such as using CPU Lite Load with profile 8. Switching to Adaptive Core Voltage Mode could help manage voltage dynamically. For monitoring, Socket Sense or VCC reporting offers more accurate insights than the current setup.
The large puddle of TIM in the bottom left corner of the photo. It's not the worst I've seen, but it's not particularly impressive either. So the LLC configurations were fine. Return to Level 7 and increase V core until it functions (you might need 1.35-1.4V). I'd expect that outcome if LLC 2 were actually LLC 2. I haven't experimented with them, I stick to the traditional OC method where they aren't really necessary. They could help save power when idle, though that's all I can say for sure. The way the VCore reading is actually measured is through Socket Sense/VCC, with three or four different sensors on the board that can swap readings. It might also slightly alter the LLC slopes, though not significantly.