Throttle Stop 9.6 Tips 14700k
Throttle Stop 9.6 Tips 14700k
Hi,
Recently changed my 12400 for a 14700k. I only have a smaller case being a Fractal Meshify C and my new GPU literally fills it to the max being 340MM, this means I cant add a 360aio to the front of the case and the top is limited to a much smaller cooler that will probably have the same thermal performance as my Noctua NHD15 Chromax Black. I will eventually upgrade my case and grab a z790 board with DDR 5 ram. Just dropped £1000 on pc parts so getting them right now is not happening, I really hope my board is not causing issues I am not bothered about overclocking this chip as it seems pretty overclocked as it is stock.
Currently I have an MSI Mag Mortar B660m Wi-Fi with 3600mhz DDR4 ram flashed to the latest bios.
I noticed thermal issues straight away with this chip hitting upwards of 85c in games, I decided to remount the cooler and it didn't help much at all. I have a contact frame arriving today which I will install as soon as it gets here.
I have for the first time been messing around with undervolting the CPU to lower the temps and stop the thermal throttling in Cinebench r23. Default my board sets the power limit to maximum, I managed to find the voltage protection setting within the MSI microcode after hours of thinking I could not undervolt on this board, then I had to disable Intel Core Isolation to get the undervolt to stick and apply.
I tried the MSI Lite Mode settings but anything lower than the auto preset of 9 just kills the Cinebench score, setting 5 halves the score and lower gets even worse. I have left this on Auto 9. Everything else in the bios is default except disabling the voltage protection.
In ThermalStop 9.6 I have set the TPL 1 + 2 both to 253 with Sync MMIO ticked and both TPL's Clamped.
CPU Core Offset Voltage is set to -125.0
CPU P Cache -90.8
CPU E Cache - 90.8
ICCMax is 511.75
My ratios are the defaults 56 56 55 recurring on the P cores and 43 on the E cores
Although I am getting better temps and scores now, I am wondering why ThrottleStop shows a CPU Mhz of 5.3ghz during multicore testing and HWInfo/Task Manager show an average core of 4.6-4.8ghz. Am I doing something wrong?
HW Info shows Package/Ring Power Limit exceeded as YES
Core Power Limit Exceeded YES
Any help/advice would be appreciated.
Kinda regretting getting this CPU at the moment and should have got the 7800X3D instead to go with the 6950xt I picked up at the same time.
Temps wise atm it idles around 30c on the core with 35-37 on the package with random spikes. Gaming before the underclock and only using the box cooler setting which essentially sets the 253w limit dropped me down to around 75c highs in games. Have not tested games with the undervolt yet,
Thanks!
Edit: Current Cinebench R23 multicore score is around 35100.
Share some ThrottleStop images that display your configuration. Include the main interface, along with the FIVR and TPL windows. I like to verify the MMIO Lock setting, which turns off the extra turbo power restrictions. These MMIO limits aren’t required. If you enable the MMIO Lock box, you won’t need to check the Sync MMIO option either.
When using ThrottleStop, close HWiNFO and open the ThrottleStop Limit Reasons window. This will indicate when and why your CPU is reducing performance. HWiNFO removes throttling information from the processor as it gathers data, which is why it’s advisable not to run both programs simultaneously when reviewing reasons for throttling.
Post a ThrottleStop screenshot that highlights this setting. ThrottleStop relies on high-performance timers inside the CPU and follows Intel’s suggested monitoring approach. This ensures accurate tracking of the CPU multiplier and MHz values. Most other tools fail to capture these details precisely.
This issue becomes clear when the CPU is throttling—whether due to thermal limits, power caps, or current restrictions. ThrottleStop avoids these constraints entirely. The Intel recommended monitoring method it uses can detect even minor changes in the CPU core multiplier accurately. HWiNFO often underestimates the MHz during throttling.
Improved scores confirm that the MHz reported by ThrottleStop reflects your actual CPU performance.
CPU Monkey has an average rating of 33572 for a 14700K, while Anandtech scored 34440. You appear to be outperforming the norm.
Edit – Here’s an example from a 10850K test in Cinebench. PL1 and PL2 were set to 150W, triggering power limit throttling. ThrottleStop displays all 20 threads operating at a multiplier just over 33. HWiNFO fails to accurately report individual core clocks or average clock speeds during this restriction.
When every 10 cores and 20 threads are engaged, they synchronize and use the same multiplier. For these processors, it’s impossible for one core to run at 1700 MHz while another is at 1600, 1800, and the rest at 5000 MHz. HWiNFO’s reported Core Clock averages greatly exceed what each core and thread actually shows.
What stands out about ThrottleStop data is its clear indication of CPU throttling over time. It can’t capture all threads simultaneously in a single measurement. ThrottleStop samples each thread individually, which causes the CPU to appear slower during that brief window. Here are the average multipliers for each core:
Core 0 – 33.215
Core 1 – 33.195
Core 2 – 33.185
Core 3 – 33.175
Core 4 – 33.170
Core 5 – 33.135
Core 6 – 33.090
Core 7 – 33.090
Core 8 – 33.080
Core 9 – 33.055
The gradual drop in C0% from 98.9% to 98.7% further confirms increasing throttling. There’s no other Intel monitoring tool that matches this level of detail.
Thank you for your kind words and the detailed update. I’ve just set up the contact frame and adjusted the core offset voltage to 90.8, matching the e and p cores after experiencing system hangs and crashes when the core reached 125. I’m currently working within that limit to avoid issues, though a crash is still possible. This might still cause problems, but it’s progressing well.
Main Window:
Voltage FIVR Control Window:
TPL Window:
Cinebench Running + Limits Window:
Have you included everything needed for analysis? Thank you again for your support.
It's not very obvious from what you're saying whether you're attempting to run your CPU overclocked or not.
Since you are doing it, having the multipliers at their maximum and lowering the vcore only extends how long the CPU runs those multipliers, rather than lowering temperatures and boosting performance.
If you aim for better temperatures, set PL1 lower than PL2 and even cut PL2 down to what your cooler can handle.
Currently everything is at full power and you're puzzled about why it's still running so high.
Everything is set to default, I haven't adjusted or changed the clocks at all. My motherboard doesn't support overclocking because it's a B660M model, not the Max version or z690/790.
I've capped the Short and Long Boost (TPL) at 253 and lowered the voltage offset for the Core, P, and E Cache to -0.0908.
I don't think I've ever overclocked before—I wasn't doing it back then either. My goal is to lower temperatures while maintaining similar or better performance.
I'm new to under-volting and haven't tried overclocking since using a 4790k and a 965 Phenom BE when I was younger.
Thanks for the advice.
During initial testing, certain users with comparable 13th Gen CPUs maintained the CPU E Cache voltage at +0.0000. They only reduced the voltage for the CPU Core and CPU P Cache. Among these, the P Cache proved crucial for stability.
Adjusting some software, such as Cinebench, by increasing the Core offset to a larger value compared to the cache offset could yield improved performance. However, this may not significantly impact results. I haven’t personally tested a 14700K, but this is something to consider. Keeping the P Cache stable at -90 mV allows you to safely raise the Core offset to around -120 mV if needed. If this leads to better outcomes, that’s ideal. If not, maintaining both offsets equally might be sufficient.
During testing, keep the E Cache voltage at +0.0000 to avoid interference. Once you stabilize the key voltages, experiment with adjusting the E Cache and observe any changes in performance or temperature. Running Cinebench consistently will help you gauge improvements or lack thereof. It seems the stability of the P Cache and E Cache is more important than equalizing both values.
Optimizing these settings can enhance stability and overall results. For instance, setting the E Cache at -150 mV or -200 mV might deliver better performance. ThrottleStop is functioning well; it confirms that PL2 power limits are correctly applied at 253 W.
Understanding multipliers like 52.16 is essential. These values reflect internal adjustments the CPU makes to balance speed and power consumption. When under heavy load, the CPU rapidly modifies these multipliers hundreds of times per second, which can hinder performance if not properly managed.
The goal remains to maximize speed without exceeding power limits. Since the CPU uses whole-number multipliers, it seeks the highest feasible value before hitting 253 W. Adjusting turbo power limits—such as increasing them to 275W or 300W—can help maintain performance during stress tests like Cinebench. Using ThrottleStop during testing provides real-time feedback on how changes affect speed and temperatures.
These power limits don’t need to be equal. Intel’s historical recommendations suggest PL2 should be 20% to 25% higher than PL1, with a turbo time limit of about 28 seconds. You might consider raising PL2 to 275W or 300W if you experience thermal throttling after 8 seconds at full power. This adjustment can help achieve peak performance for short bursts while minimizing power restrictions.
Thank you for sharing your findings. It’s encouraging to see how a 14700K performs on the B660 board. The Mortar series boards appear to offer strong cooling for their voltage regulators, whereas many low-end models struggle with thermal management. Running a 14700K at full power on less efficient boards often triggers warning lights for VR temperature or current, indicating overheating concerns.
There’s no overclocking capability on a B660 board. The real need is to disable Undervolt Protection, undervolt the CPU, and fine-tune turbo power limits. The default voltage curve tends to be too high, so lowering it can significantly improve efficiency and performance. Reducing power consumption directly translates to better heat management and overall speed.
You appreciated the response; however, I've been quite occupied and didn't have time to fine-tune the CPU further. I've been experimenting today.
I experienced a crash or system freeze when all three settings were at 90.8 during a game of Overwatch.
I adjusted the E Cache to the default value of 0 as you recommended, lowered the P Cache offset to -70.3, and increased the Core Voltage offset to -140.6.
This adjustment enabled the cores to operate just over 5.4Mhz in Throttlestop during Cinebench testing, and so far it appears stable. However, I'm unsure which software to use for optimal stability testing. I've noticed occasional flickering red from one of the two EDP components for a brief moment, even when idle. Should I be worried about these warnings?
Setting the TPLs above 253 doesn't seem to significantly affect frequency, and adjusting the E Cache voltage offset didn't help either. Resetting all voltages and keeping TPL at 253 still causes thermal throttling well before reaching the 100°C limit in my BIOS.
Interestingly, I can reach 93°C without any voltages set and still experience throttling, but when I raise the TPL above 253 with voltages set, I don't throttle at the same temperatures as without voltages, yet I haven't seen any improvements in frequency or score.
Currently, I'm achieving over 35k in Cinebench. I'm not entirely confident about stability, as repeated crashes across all three cores at 90.8 took a long time and often led to a full system shutdown requiring me to power off via the case button.
Regarding temperatures, lowering them further doesn't seem to yield major improvements; it appears to peak around 88°C with most voltage settings I've tried.
Today's BIOS update included several changes:
- Updated CPU uCode.
- Enhanced Secure Boot mechanism.
- Revised OC Auto rules.
- Optimized Intel APO function.
- Firmware version: ME_16.1.30.2264 (download).
- ME Firmware update SOP.
I was already running the ME FW version.
I did encounter a system hang at -160 Core, -70 P Cache, and 0 E Cache during Cinebench testing. I've run Cinebench over 60 cycles with these settings and 253 TPLs.
Edit: I also experimented with Turbo Time and left it at 56; 28 and 8 settings didn't seem effective.
Should I enable the Speed Shift EPP 128 option in BIOS as Auto but not in Throttlestop? Speedstep is enabled by default and is already ticked there.
I saw a video suggesting Intel Dynamic Tuning Technology for the 14700k, which was disabled by default in my BIOS. I've now enabled it, though I'm not sure if it will help.