Undervolting and Overclocking

Undervolting is the process of reducing the operating voltage (Vcore for CPUs and Vddc for GPUs) Supplied below the voltage level defined by the manufacturer at a given clock speeds.

Reducing power consumption ($P = C \times V^2 \times F$), reducing heat generation due to the lower Voltage supplied, and hence reducing cooling required.

1. Update BIOS/UEFI to latest stable version
2. Backup BIOS profile
3. Install monitoring tool (HWiNFO64) and stress testers (Cinebench R23, y-cruncher, MemTest86 and Prime95)
4. Press Del on boot when the logo can be seen to enter BIOS
5. Enter advance mode, this can be different depending on the manufacture
6. Find Core Voltage Offset and set to $-0.075V$
7. Save BIOS changes (F10)
8. If the system fail to boot
9. Wait for 3 auto recovery attempts
10. Clear CMOS by powering off, removing the battery for 60s or short JBAT1 pins
11. If the system successfully boot, run the following test, any failure, change the voltage offset by $0.010 V$
12. Smoke test with cinebench R23 multicore
13. Run HWiNFO with option sensor only, and click logging start then follow the instructions.
14. Run y-chruncher with VST for 15 minuites
15. Run Prime95 for 60 minutes, setting Test=FFT,MinFFT=4,MaxFFT=4096,RunFFT=1,Memory=90,Time=60
16. (Optional) Test the idel stability by Start-Sleep -Seconds 7200 in powershell for 2 hours.
17. Check HWiNFO logs for WHEA errors, if any increase voltage.
18. If no problem arrises, increase the core voltage offset by $0.010V$ again for stability

1. Install MSI Afterburner
2. Open the voltage/frequency curve editor (Ctrl+F)
3. Find the voltage point corresponding to your desired clock speed
4. Drag the voltage point down (lower voltage) while keeping the frequency the same
5. Apply and test with a GPU benchmark (Superposition, 3DMark Time Spy)
6. If stable, continue lowering voltage in 10–25 mV steps until instability appears
7. Raise voltage by one step from the last unstable point
8. Verify stability with a 30-minute FurMark session and monitor temperatures

Understanding voltage is foundational to all tuning. This section covers the key concepts that apply To both undervolting and overclocking.

Vcore is the actual voltage delivered to the CPU cores. VID (Voltage ID) is the voltage the CPU requests from the Integrated Voltage Regulator (IVR). These two values are not always the same:

• At idle: Vcore may be close to or slightly below VID due to power-saving states (C-states).
• Under load: Vcore is below VID due to Vdroop.
• The delta between VID and Vcore is affected by LLC (Load-Line Calibration) settings.

When you set a voltage offset of -75 mV, you are telling the VRM to deliver 75 mV less than the VID The CPU requests. The actual Vcore will be even lower due to Vdroop.

Vdroop is an intentional design feature of CPU voltage regulators. When the CPU load suddenly drops (for example, a compute thread finishes), the inductance in the VRM circuit would cause the voltage To overshoot if there were no droop. This overshoot can damage silicon. Vdroop provides a margin so That the worst-case overshoot stays within safe limits.

The relationship is simple: higher load → more Vdroop → lower actual voltage. This is why your CPU May be stable at 1.25 V under light load but crash under heavy load even though the VID has not Changed.

LLC counteracts Vdroop by making the VRM more aggressive in maintaining the target voltage under Load. LLC is configured on a scale (e.g., Level 1–7, or Auto/Medium/High/Extreme).

Warning: Setting LLC too high eliminates the safety margin that Vdroop provides. On load Transitions (load → idle), the voltage can overshoot the VID by a significant margin. This is Particularly dangerous for CPUs running near their maximum safe voltage. For most use cases, LLC Level 4 or 5 provides a good balance.

Exceeding these limits can cause electromigration and permanent silicon degradation:

These are conservative long-term limits. Short bursts above these values (transient spikes) are Normal and handled by the CPU”s internal protections. The concern is sustained voltage at or above These thresholds under load.

ThrottleStop is the preferred tool for undervolting Intel laptops. It works by directly Communicating with the CPU’s Integrated Voltage Regulator (IVR) to apply voltage offsets.

1. Download ThrottleStop from the official thread on NotebookReview forums.
2. Extract and run as Administrator.
3. On the “FIVR” (Fully Integrated Voltage Regulator) screen, click “Unlock Adjustable Voltage.”

1. In the FIVR control window, find the “CPU Core Voltage Offset” slider.
2. Set a negative offset. Start with -50 mV for a safe starting point.
3. Click “Apply” and immediately run Cinebench R23 multi-core.
4. If stable, reduce the offset by an additional 10 mV (e.g., -60 mV).
5. Repeat until the system crashes or shows artifacts.
6. When instability occurs, increase the offset by 10 mV from the last unstable point.

On 12th Gen and later Intel CPUs, P-Cores (Performance) and E-Cores (Efficiency) have separate Voltage domains. ThrottleStop allows you to set independent offsets for each:

• P-Core Offset: The P-Cores do the heavy lifting and need less aggressive undervolts. Start at -50 mV.
• E-Core Offset: The E-Cores are more tolerant of lower voltage. You may be able to push these to -80 mV or beyond.

• Disable Speed Shift (SpeedStep Technology = Off): Some users report more consistent undervolting behavior with Speed Shift disabled. This prevents the CPU from rapidly changing P-states, which can cause voltage transitions that trigger instability.
• Set Power Limits: ThrottleStop can override the OEM’s PL1 and PL2 power limits. This is useful when the laptop manufacturer has set overly conservative limits.
• Enable “BD PROCHOT”: Controls the external PROCHOT signal. Disabling this can prevent the laptop from throttling when the charger gets warm, but it also removes a safety mechanism.

Intel XTU provides similar functionality with a more polished GUI but is less reliable on laptops. It is better suited for desktop platforms. The same offset approach applies: start at -50 mV, reduce In 10 mV steps, stress test after each change.

Intel has locked undervolting on many 12th, 13th, and 14th Gen desktop platforms via microcode Updates. Check your BIOS for a “Microcode Update” or “CPU Undervolting Protection” setting. On some Motherboards, this can be disabled to restore undervolting capability. On laptops, ThrottleStop may Still work depending on the OEM’s implementation.

AMD’s approach to undervolting is fundamentally different from Intel’s. Instead of applying a fixed Voltage offset, AMD uses Curve Optimizer to shift the entire voltage-frequency curve.

The Curve Optimizer applies a signed offset (positive or negative) to the voltage requested by the CPU’s boost algorithm at each frequency point. A negative offset means the CPU tries to hit the same Frequencies at lower voltage.

• Negative values (e.g., -20) = undervolting (lower voltage, same frequency)
• Positive values (e.g., +20) = overvolting (higher voltage, same frequency) — used to support higher overclocks
• Units: The Curve Optimizer is measured in steps, where each step is approximately 5 mV. A value of -20 corresponds to roughly -100 mV.

1. Enter BIOS and navigate to AMD overclocking settings ( under “Advanced” or “AMD CBS”).
2. Find “Curve Optimizer” or “Precision Boost Overdrive → Curve Optimizer.”
3. Enable Curve Optimizer and select “All Core” or “Per Core.”
4. For an all-core undervolt, start with -15 (approximately -75 mV).
5. Save, reboot, and stress test.
6. If stable, reduce by 2 steps (e.g., -17).
7. If unstable, increase by 2 steps.

AMD CPUs have a “Core Quality” ranking that indicates which cores are stronger. You can apply more Aggressive negative offsets to higher-quality cores and less aggressive offsets to weaker ones.

1. Use Ryzen Master or HWiNFO64 to identify the best and worst cores.
2. Apply the most aggressive negative offset (e.g., -20) to the best cores.
3. Apply a milder offset (e.g., -10) to the weakest cores.
4. Test each core individually using single-thread workloads.

This is time-consuming but can yield better results than a uniform all-core offset.

AMD’s power limits control how much power the CPU is allowed to draw:

For undervolting, you do not need to change these. However, if you are trying to reduce Thermals on a laptop or SFF build, lowering the PPT can force the CPU to boost less aggressively, Which reduces temperatures at the cost of peak performance.

• Stock PPT values: Varies by SKU. A Ryzen 9 7950X has a 170 W TDP with 230 W PPT.
• Reduced PPT example: Setting PPT to 120 W on a 170 W CPU will reduce multi-core performance by 10–20% but significantly improve thermals.

Enabling PBO allows the CPU to boost beyond its rated limits. Combined with a negative Curve Optimizer offset, this can yield higher sustained frequencies at lower temperatures than stock.

1. Enable PBO in BIOS.
2. Set PBO limits to “Manual.”
3. Set PPT to “Motherboard” (lets the board’s VRM determine the limit) or a custom value.
4. Apply Curve Optimizer with a negative offset.
5. Stress test and monitor.

Warning: Enabling PBO without a Curve Optimizer offset can increase temperatures significantly. The recommended approach is to undervolt first, then enable PBO.

CPU overclocking increases the clock speed beyond the manufacturer’s rated specifications. This can Be done by adjusting the base clock (BCLK), multiplier, or both.

Modern CPUs have two ways to increase frequency:

• Multiplier: The CPU’s internal clock multiplier. On unlocked CPUs (Intel K-series, AMD Black Edition / Ryzen), this is freely adjustable. Changing the multiplier only affects CPU frequency.
• BCLK (Base Clock): The reference clock that drives the CPU, memory, PCIe, and DMI/UPI interconnects. Increasing BCLK overclocks everything simultaneously. This is more complex and risky because it can destabilize PCIe devices, storage controllers, and memory.

Recommendation: Always use multiplier overclocking when possible. BCLK overclocking should only Be attempted when you have exhausted multiplier headroom and understand the risks to other Subsystems.

1. Enter BIOS and enable “Overclocking” or “XMP” mode (required to unlock multiplier settings on some boards).
2. Set all-core multiplier to a target value (e.g., 50x for 5.0 GHz).
3. Set a fixed Vcore or adaptive voltage:

• Fixed Vcore: Set manually (e.g., 1.28 V). Simple but wastes power at idle.
• Adaptive Voltage: Set an offset or target voltage with a negative offset. More power efficient but harder to stabilize.

4. Set LLC to Level 4–5.
5. Save, boot, and stress test with Prime95 Small FFTs.
6. If stable for 30 minutes, try increasing the multiplier by 1.
7. If unstable, increase Vcore by 0.01–0.02 V.
8. Repeat until you reach your temperature or voltage limit.

Target temperatures: Stay below 90 °C under Prime95 Small FFTs. If you exceed 90 °C, you have Hit your thermal limit and should stop increasing frequency.

AMD’s boosting algorithm (Precision Boost 2 / Precision Boost 2 with Curve Optimizer) makes manual Overclocking less relevant on Zen 3 and later. The algorithm already pushes the CPU close to its Maximum frequency. Manual overclocking disables PBO and can actually reduce performance in some Workloads.

For Zen 2 / Zen 3 / Zen 4:

1. The recommended approach is to enable PBO with a negative Curve Optimizer offset rather than setting a fixed frequency.
2. If you insist on manual overclocking, disable PBO and set a fixed frequency and voltage.
3. Start with a conservative target (e.g., 100 MHz above stock all-core boost).
4. Set a fixed Vcore and LLC level.
5. Stress test with y-cruncher or OCCT.
6. Increment frequency and voltage as needed.

BCLK overclocking is an advanced technique that increases the reference clock frequency:

• Default BCLK: 100 MHz
• Typical target: 102–105 MHz
• Aggressive target: 105–110 MHz (high risk of instability)

What BCLK affects:

• CPU frequency (multiplier × BCLK)
• Memory frequency (ratio × BCLK)
• PCIe and DMI/UPI interconnect speeds

Risks:

• PCIe devices (GPU, NVMe SSD, network cards) can become unstable if the BCLK is too high.
• NVMe SSDs can corrupt data if the PCIe link becomes unstable. Back up your data before attempting BCLK overclocking.
• Some platforms (Intel 12th Gen+) have a fixed BCLK and do not support BCLK overclocking.

Stop and revert to your last stable settings if any of the following occur:

• Core temperatures exceed 95 °C under sustained load
• You are approaching the maximum safe Vcore for your platform
• WHEA errors appear in HWiNFO64 logs
• The system requires significantly more voltage for marginal frequency gains (diminishing returns)
• You experience random reboots, BSODs, or application crashes during normal use
• Your VRM temperatures exceed 80 °C (check with HWiNFO64)

Memory overclocking increases the data rate and/or tightens the timings of your RAM. This can Improve performance in memory-bound workloads (gaming, compilation, scientific computing).

XMP (Extreme Memory Profile) and EXPO (Extended Profiles for Overclocking) are vendor-certified Overclocking profiles stored in the RAM’s SPD (Serial Presence Detect) chip. Enabling XMP/EXPO in BIOS applies the manufacturer’s tested settings.

1. Enter BIOS.
2. Navigate to memory settings.
3. Enable XMP (Intel) or EXPO (AMD).
4. Select the desired profile (Profile 1 or Profile 2 if available).
5. Save and reboot.

Important: XMP/EXPO profiles are certified by the RAM manufacturer for use with a single module. Running two or four modules at XMP speeds is not guaranteed to be stable, as the memory controller Must work harder to drive multiple modules. If you experience instability with multiple modules, Try:

• Reducing the frequency by one step (e.g., from 6000 MT/s to 5600 MT/s)
• Relaxing the timings
• Increasing the DRAM voltage by 0.05 V

Memory timings are specified as a series of numbers (e.g., 16-18-18-38). These represent delays Measured in clock cycles:

Lower numbers = better performance. Tightening timings is more difficult than increasing frequency But can yield similar or better performance gains in latency-sensitive workloads.

Beyond the primary timings (CL-tRCD-tRP-tRAS), there are secondary and tertiary timings that affect Performance:

• tRC (Row Cycle Time): Related to tRAS + tRP. Often auto-calculated but worth tightening manually.
• tFAW (Four Activate Window): Minimum time between four activate commands. Important for four-DIMM configurations.
• tRFC (Refresh Cycle Time): Time to complete a DRAM refresh cycle. Very long on DDR5; can often be tightened significantly.
• tREFI (Refresh Interval): Time between refresh cycles. Increasing this improves performance but may cause data errors if set too high.

• DDR4: Mature platform, extensive tuning guides available. Sweet spot for Ryzen 3000/5000 is 3600 MT/s with tight timings (CL14–CL16).
• DDR5: Higher default speeds but much looser timings. The memory controller on AMD Zen 4 and Intel 13th/14th Gen handles DDR5 well up to 6000–6400 MT/s. Beyond 6400 MT/s, the memory controller becomes the bottleneck, especially with two DIMMs populated.

After any memory frequency or timing change, run these tests:

1. MemTest86 (bootable): Run at least 4 passes. This tests memory without OS interference.
2. TestMem5 / TM5 (Windows): Run the “Extreme by Anta777” or “1usmus” config for 30–60 minutes. Catches timing instability that MemTest86 may miss.
3. Karhu RAM Test (Windows, paid): The most thorough Windows-based memory tester. Run for at least 3–4 cycles. Expensive but worth it if you are serious about memory tuning.
4. y-cruncher: Run a memory-intensive test (e.g., Pi computation). Catches instability under real workloads.

Stress testing is not optional — it is the only way to verify that your tuning changes are stable. A System that “feels fine” during normal use can still be producing WHEA errors that indicate silent Data corruption.

Follow this order after any tuning change:

1. Quick smoke test: Cinebench R23 multi-core, 1 run. Takes 10 minutes. Catches gross instability.
2. Short stress test: y-cruncher VST, 15 minutes. Catches moderate instability.
3. Extended stress test: Prime95 Small FFTs, 60 minutes. Catches thermal and voltage instability.
4. Memory test: MemTest86, 4+ passes. Catches memory-specific instability.
5. Idle test: Leave the system idle for 2 hours with HWiNFO64 logging. Catches C-state instability (system crashes or freezes when idle).

For CPU stability testing, use these settings:

Test=FFT, MinFFT=4, MaxFFT=4096, RunFFT=1, Memory=90, Time=60

• Small FFTs (FFT=4–4096): Stresses the CPU with data that fits in L1/L2 cache. Generates maximum heat. Use this for thermal testing.
• Blend (large FFTs, high memory): Stresses both CPU and memory. Use this for overall system stability.
• Run time: 60 minutes is the minimum. For production systems, run overnight (8+ hours).

OCCT is particularly good at catching instability that Prime95 misses:

• OCCT CPU (Small Data Set, AUTO): Similar to Prime95 Small FFTs but uses a different stress algorithm. Run for 30 minutes.
• OCCT CPU (Large Data Set): Tests with data larger than L3 cache, stressing the memory controller. Run for 30 minutes.
• OCCT Memory: Dedicated memory stress test. Run for 60 minutes after any timing or frequency change.

Run HWiNFO64 in “Sensors Only” mode with logging enabled. Key sensors to watch:

• CPU Package Temperature: Should stay below 90 °C under sustained load.
• CPU Core Max: The hottest individual core. More useful than package average.
• CPU Vcore: Actual delivered voltage. Should be stable (not fluctuating wildly) under load.
• CPU VID: Requested voltage. The delta between VID and Vcore indicates Vdroop.
• CPU Package Power: Total power draw. Compare against your PPT/PL2 limit.
• VRM Temperature: Voltage regulator temperature. Should stay below 80 °C.
• Memory Temperature: For DDR5, keep below 60 °C. DDR4 is generally safe below 50 °C.
• WHEA Errors: Any non-zero count indicates instability.

For real-time monitoring during benchmarks and games:

1. Install RivaTuner Statistics Server (included with MSI Afterburner).
2. Configure the OSD to display CPU temperature, GPU temperature, CPU power, GPU power, frame rate, and 1% low frame times.
3. Toggle OSD with the hotkey (default: Ctrl+Shift+O).

If temperatures are too high, address these in order:

1. Repaste the CPU/GPU. Thermal paste degrades over time. Replacing it with a quality paste (Thermal Grizzly Kryonaut, NT-H2, or SYY-157) can drop temperatures by 5–15 °C.
2. Improve case airflow. Ensure intake fans provide cool air and exhaust fans remove hot air. A single fan configuration is rarely adequate for tuned systems.
3. Undervolt. This is the most effective way to reduce temperatures without sacrificing performance.
4. Reduce power limits. Lowering PPT/PL2 reduces the maximum power draw, which directly reduces temperatures.
5. Upgrade cooling. A better cooler (larger tower heatsink, liquid cooling, or a laptop cooling pad) may be necessary if the above steps are insufficient.

Every silicon sample is different. Settings that are stable on one CPU may crash another of the Exact same model. Do not assume that someone else’s settings will work for you. Always test Thoroughly with your own hardware.

Diminishing returns are severe near the limit. Going from 5.0 GHz to 5.1 GHz might require 0.05 V More voltage and increase temperatures by 5 °C, for a 1–2% performance gain. Going from 4.8 GHz to 5.0 GHz might cost 0.02 V and yield a 3–5% gain. Focus on the low-hanging fruit: undervolting for Sustained boost, enabling XMP/EXPO, and ensuring adequate cooling.

WHEA errors are not harmless. They indicate that the CPU detected and corrected bit flips. While the CPU can correct single-bit errors, multi-bit errors cause uncorrectable machine check exceptions (UCMCE), which result in BSODs. More importantly, silent data corruption can occur before the error Is detected. If HWiNFO64 reports any WHEA errors during stress testing, your settings are not Stable.

Some instability manifests only during load transitions (load → idle, idle → load). This is Particularly common with aggressive LLC settings and fixed voltage modes. Always test idle stability After confirming load stability.

Before making any change, save your current BIOS profile. If something goes wrong, you can quickly Restore to a known-good state without having to clear CMOS and reconfigure everything from scratch.

Motherboard manufacturers frequently release BIOS updates that change power delivery behavior, add New tuning options, or fix stability issues. An overclock that was stable on BIOS version 1.20 may Be unstable on 1.30. After any BIOS update, re-test your tuning settings.

Your PSU must be able to deliver the power your components demand. An overclocked CPU and GPU can Draw significantly more power than stock. If your PSU is operating near its capacity, voltage ripple Can increase, which causes instability that is not related to your CPU or GPU settings.

Buying two separate RAM kits (even of the same model and speed) and using them together is a common Source of memory instability. RAM kits are binned and matched at the factory. Two kits from Different batches may have different memory ICs or require different voltages. Always buy a single Kit with the total capacity you need.

NVIDIA GPUs (RTX 20-series and later) use a voltage/frequency curve that can be edited with MSI Afterburner:

1. Open MSI Afterburner and press Ctrl+F to open the curve editor.
2. Hold Shift and click on the voltage/frequency curve to create a custom point.
3. Find the clock speed you want to sustain (e.g., 1860 MHz).
4. Set the voltage to the lowest point on the curve that supports that clock speed.
5. Drag all points above your target voltage down to the same voltage level, creating a flat line.
6. This ensures the GPU never exceeds your target voltage, regardless of load.

Typical results:

• RTX 3060: 1800–1900 MHz at 825–875 mV (stock is ~1050 mV)
• RTX 3070: 1800–1900 MHz at 875–925 mV
• RTX 4070: 2500–2600 MHz at 875–925 mV

1. After undervolting (to establish a thermal baseline), increase the core clock offset in 10–15 MHz increments.
2. Test with Superposition or 3DMark Time Spy after each increment.
3. When artifacts or crashes appear, reduce the offset by 15–20 MHz.
4. Optionally, increase the memory clock offset in 50–100 MHz increments.
5. Memory overclocking responds differently. Test with a VRAM-intensive benchmark.

AMD GPUs can be tuned with MSI Afterburner or AMD’s own Adrenalin software:

• Adrenalin Undervolting: Use the “Tuning” tab → “Automatic Undervolting” or set a manual curve. The interface is less flexible than MSI Afterburner.
• Power Limit: AMD GPUs have a power limit slider that can be increased by 10–15%. Combined with undervolting, this can yield higher sustained clocks at lower temperatures.

Knowing when to stop is more important than knowing how to push further. Stop tuning and enjoy your System when:

• Your temperatures are comfortable (below 80 °C under sustained gaming/workload)
• You have no WHEA errors after extended stress testing
• The performance gains from further tuning are marginal (< 2%)
• You have reached a point where each additional step requires disproportionately more voltage or cooling
• You are no longer enjoying the process — tuning should be fun, not frustrating

The best tune is the one that gives you the performance you need with comfortable temperatures, low Noise, and rock-solid stability. Everything beyond that is diminishing returns.

This topic covers the essential concepts and techniques related to undervolting and overclocking, including key principles and practical applications.

Key concepts include:

• core concepts and definitions
• key principles and frameworks
• practical applications
• common techniques and methods
• evaluation and critical analysis

A thorough understanding of these concepts, combined with regular practice and review, is essential for mastery of this topic.

Worked examples demonstrating the application of key concepts are covered in the detailed sub-pages linked above.