Search

Technical details and guidance on Undervolting

Introduction

Undervolting refers to the deliberate reduction of the electrical voltage supplied to the CPU and GPU. When applied correctly, undervolting can slightly reduce system temperatures and power consumption without compromising system stability. Under high load, undervolting can also improve overall performance, as it delays the point at which temperature limits are reached, giving the voltage-reduced components more time to reach their full boost potential.

Undervolting is theoretically also possible for graphics cards, but it is significantly more complex and error-prone. This article therefore deals exclusively with CPU undervolting.

Methodology: run-time vs. boot-time

There are essentially two methods for changing the "voltage offset" value:

  • Via the BIOS setup directly at system startup (boot time)

  • Via Windows software during operation (run time)

Both methods have different advantages and disadvantages.

  Run-time Voltage Offset Boot-time Voltage Offset
Advantages Does not require reboot.

No risk of “no boot” scenario.		 Once it is set, it will stick (unless you reset or update BIOS). “Fire & Forget” principle.
Disadvantages Does not work when VBS (Core Isolation, Memory Integrity, Hyper-V) is enabled.

Can be abused by malware to steal secrets (e.g. break into password manager’s memory space). Could be accidentally reset back to default.		 Requires reboot to change settings.

Risk to cause “no boot” scenario with extreme settings (requires recovery method/hotkey).

The method supported by each model is specified in the overview table linked above.

VBS blocks run-time undervolting

The "Memory Integrity" option in the Windows security settings has been enabled by default since Windows 11. This option is also known as VBS (Virtualisation-based Security) or HVCI (Hypervisor-protected Code Integrity).

If this function is active, CPU undervolting via the so-called run-time method is generally not possible, as VBS prevents software from accessing security-relevant CPU registers during operation.

win11_vbs_memory-integrity_on_en.png
Security → Device security → Core isolation → Memory integrity

It is possible to disable Memory Integrity in order to allow run-time undervolting. However, this reduces the security level of the system, as protection mechanisms against certain classes of malware and kernel exploits are disabled. Such a change should therefore only be made consciously and after carefully weighing the associated risks.

Potential security risk due to Plundervolt

A system that supports the run-time method described above, or makes it accessible by disabling VBS, also carries an additional risk: the method could be exploited by malware to extract secrets from protected memory areas, as demonstrated by the Plundervolt exploit.

It is irrelevant whether the user has already performed undervolting themselves. The risk does not lie in the user’s application of undervolting, but solely in whether run-time voltage offset is provided or allowed by the system.

Recovery method to prevent a "no boot" situation

As explained in the table above, the boot-time method carries the fundamental risk that excessive undervolting can prevent the system from booting. In such a case, the BIOS setup can no longer be accessed to reset the excessive values. This is referred to as a "no boot" situation. Recovering from this state requires resetting the tuning values using a workaround. A simple CMOS reset, for example by removing the CMOS battery, is not sufficient.

Automatic recovery

To mitigate this issue, we introduced a comprehensive recovery method in the XMG CORE and XMG NEO series as early as 2021. The EC microcontroller (EC = Embedded Controller), which operates independently of the CPU, detects whether the system has failed to boot several times in succession. If this is detected, the EC firmware ensures that the next boot process is carried out with default values, allowing the user to re-enter the BIOS setup and adjust the settings. In some model series, the EC firmware also performs a direct BIOS reset.

Manual recovery via Ctrl+B

This recovery process can also be triggered manually using the Ctrl+B hotkey. If you hold down this key combination during a cold start, the tuning options are reset manually. You should then return to the BIOS setup to make further adjustments.

xmg-neo-15_ctrg-b-hotkey.jpg
Pressing Ctrl+B with the left hand on XMG NEO 15

Important:

  • Recovery with Ctrl+B only works during a cold start, not on restart.
  • If the unit is in a reboot loop due to non-bootable settings, it must first be switched off completely by pressing the power button for a particularly long time. It is only switched off when all LEDs have gone out.
  • Before the cold boot, it is necessary to unplug the power adapter and plug it back in again shortly afterwards.

Step-by-Step Instructions:

  • Switch off the system completely by pressing and holding the power button. The system is only shut down when all LEDs have gone out.
  • Pull out the power adapter and plug it back in again a few moments later.
  • Hold down Ctrl+B, press the power button, but do not release Ctrl+B yet.
  • The recovery is then successfully completed as soon as the display backlight is visible. At this point you can release Ctrl+B and immediately start holding F2 to enter the BIOS setup.

Length of the recovery procedure depending on the model series

The length of time you have to hold down Ctrl+B or how long you have to wait afterwards depends on the model series.

Model series Length of the recovery procedure
Models up to 2021 Ctrl+B must be held down continuously for up to two minutes. As soon as the reset has been carried out successfully, the LED of the power button flashes briefly. You can then release Ctrl+B. Depending on the model series, it can now take up to 30 seconds until the next boot process takes place. In the meantime, you can press F2 to enter the BIOS setup during the next boot process.
Newer models If the recovery procedure with Ctrl+B is used correctly (cold start with previously disconnected and reconnected power supply), the procedure only takes a few seconds until the boot logo is visible again.

Safety and stability

CPU undervolting can affect system stability and, as explained above, may even prevent the system from booting. It is therefore essential to approach the optimum settings gradually and to test after each adjustment.

In terms of stability, there is a wide grey area between "does not boot" (worst case) and "occasionally causes crashes". Only after extensive testing, possibly over several weeks, can one be reasonably certain that an undervolting configuration is stable.

Undervolting is considered non-destructive

Undervolting can affect system stability or even lead to no-boot situations, but only while excessively high voltage offset values are actually applied. As soon as these values are reset to zero, all instabilities are eliminated. Undervolting therefore has no negative after-effects and does not adversely affect the service life of the device. This is a decisive advantage of undervolting compared to overvolting and overclocking. We therefore support undervolting, but not overvolting.

Stability tests

The following tests are recommended:

  • Prime95, preset "Small FFT" with AVX deactivated
  • AIDA64 System Stress Test
  • Furmark, 3DMark or Gaming Tests

Prime95 and AIDA64 by default only test the CPU. AIDA64 is considered the tougher test. Failure of such a test may result in a blue screen or an application crash. AIDA64 may also display the message "Hardware failure detected", as shown in the following screenshot:

aida64_trial-version_hardware-failure.png
Screenshot: AIDA64 "System Stability Test" mit Standard-Einstellungen (CPU, FPU, cache, system memory) und einem Fehler nach etwa 38 Minuten.

If these tests do not show any instabilities, GPU and gaming tests can also be used. Even if the graphics card is not directly affected by CPU undervolting, typical 3D load will also cause a rather high and quite fluctuating CPU load, partly due to the intensive data exchange between CPU and graphics card via the PCI Express interface.

Any kind of incidents? Dial back your undervolt!

Even if the aforementioned stability tests are successful, this does not mean that the undervolting setting is 100% stable. Instabilities can also occur sporadically with variable or low loads.

Such long-term instabilities can manifest themselves in symptoms such as software crashes, blue screens, arbitrary system shutdowns, visual image artefacts or audio glitches. In the grey area between "stable" and "unstable", this may only occur after several weeks of active use. Therefore, if unexpected incidents should ever occur, you should reduce the undervolting by a few levels and continue testing/working.

Looking back

Certain events and trends in recent years have made undervolting increasingly difficult to support, or in some cases impossible. The following chronology outlines the key developments.

Before 2019 Voltage offset was generally enabled on Intel H-series CPUs. CPU undervolting was widespread on both desktop and mobile platforms.
2019 After the Plundervolt security vulnerability was disclosed, Intel recommended that OEMs no longer offer undervolting (source). Many manufacturers followed this recommendation.
2020 It was initially unclear whether Intel’s recommendation was binding. After further consultation, Intel confirmed that voltage offset could still be offered, provided that OEMs were aware of and accepted responsibility for the associated risks. This allowed us to continue offering undervolting in many models. In this context, we published official undervolting instructions for Intel Core 10th Gen H-series CPUs on Igor’s Lab.
2021 With Intel Core 11th Gen H (Tiger Lake), we introduced an automatic recovery mechanism for XMG CORE and XMG NEO. The EC firmware resets BIOS tuning options if the system fails to boot multiple times. In addition, recovery can be triggered manually via a hotkey (Ctrl+B during a cold start). This effectively secured BIOS-based undervolting for these series.
2022

With Intel Core 12th Gen, the H series technically transitioned to the more compact P platform, which no longer includes the MSR-0x150 mailbox required for voltage offset. As a result, undervolting was no longer possible on H CPUs, with the i9-12900HK being the sole exception. At the same time, we introduced alternative tuning via AC Loadline in the BIOS for selected models.

Also in 2022, Intel released the HX series based on desktop CPUs, which continues to support voltage offset. From the end of 2022 onwards, Intel’s "Undervolting Protection" was gradually introduced via reference code updates, initially blocking undervolting entirely (Sources/examples: [1] [2] [3]).
2023 The Intel Core 13th Gen HX series generated significant interest among enthusiasts. At the same time, undervolting was initially not possible due to enabled Undervolting Protection. A problematic interaction with BIOS update mechanisms (AMI and Insyde) prevented a secure release for an extended period. After intensive coordination with Intel, BIOS vendors and ODMs, these hurdles were overcome, allowing undervolting to be officially offered again in selected models.
2024 Undervolting once again became available in most models based on the Intel HX platform. Overall public interest in the topic declined. On the AMD side, efficient mobile platforms without undervolting continued to gain traction. With AMD Dragon Range, a desktop-derived platform, undervolting became possible in principle, but without a uniform BIOS code base. With XMG APEX 17 MAX, we offered documented BIOS undervolting options on AMD for the first time.
2025 Intel replaced the 14th generation Core series with Intel Core Ultra. Undervolting remains possible only on the desktop-derived HX series, but not on the mobile H series. The same applies to AMD Fire Range, the successor to Dragon Range. Here as well, we were able to safely implement selected BIOS options despite their high complexity.

More Informationen