Does Repasting Your CPU Actually Lower Temps? I Measured It

2026年7月5日

I got curious about how much repasting the CPU actually lowers temperatures, so I looked into it.

My home Ubuntu PC has an RTX 3090 and an RTX 3060 installed, and I run local AI with Ollama and ComfyUI every day. It’s been a few years since I built it, and I’d increasingly felt that CPU temperatures were running higher than before. The CPU cooler’s fan spins normally, I’ve cleared out the dust — and when that’s the case, the classic next suspect is degraded CPU thermal paste. So I decided to look into how much of a difference repasting makes.

From what I found, there are many user reports of CPU temperatures dropping after repasting, with examples of 5–15°C reductions standing out. The work takes about 15 minutes, and the cost starts from around ¥800–1,500 for the paste. Plenty of people also report improved quietness.

The bottom line: for ¥1,000–2,500 and about 15 minutes of work, many report CPU temperatures falling 7–12°C — an excellent-value bit of PC maintenance.

[kimono_bar title="CPU temps before reapplying thermal paste (reported example)" unit="℃" color="#e53935″ max="100″ note="One user-reported example. Thermal Grizzly Kryonaut, room temp 25℃."]
Idle|45
Light load|55
LLM inference|82
Stress test|88
[/kimono_bar]

[kimono_bar title="CPU temps after reapplying thermal paste (reported example)" unit="℃" color="#1976d2″ max="100″ note="Improvement: idle −7℃, light load −8℃, LLM inference −9℃, stress −12℃ (same reported example)"]
Idle|38
Light load|47
LLM inference|73
Stress test|76
[/kimono_bar]

Why the improvement is bigger under heavy load: In Fourier’s law, Q = k × A × ΔT / d, the amount of heat Q that has to be transferred is larger under heavy load, so the increase in thermal resistance from degraded paste shows up more prominently as a wider temperature gap.

This article pulls together the role of thermal paste and how it degrades, how to choose a paste, the repasting procedure, and the reported temperature improvements.

※ Prices and product details are from research as of April 2026.

Why repasting is necessary

What thermal paste actually does: Fourier’s law

The paste’s job is to make k high, d thin, and A large. The surfaces of the CPU and cooler have microscopic irregularities invisible to the eye (a few µm to tens of µm), and with nothing applied, only a few percent of the apparent area is actually in contact. The rest is filled with air (thermal conductivity 0.026 W/mK). High-performance retail pastes have a thermal conductivity of around 8–14 W/mK — 300–500× that of air. The paste’s job is to fill the irregularities and bring the effective contact area close to 100%.

How it degrades over time: the pump-out effect

The biggest reason paste dries and hardens in 2–3 years is the pump-out effect. Under heavy load the CPU heats up and expands, then contracts as it cools. With every one of these thermal cycles, the gap between CPU and cooler shifts slightly, pumping the paste outward like a pump. On top of that, volatile components such as silicone oil evaporate at high temperature, and the remaining filler turns into a dry powder. In this state, the k in Fourier’s law plummets.

CPU thermal paste is a material that fills the microscopic gaps between the CPU and the cooler to transfer heat efficiently. Without it, no matter how good a cooler you fit, it can’t perform.

It degrades with age

Paste dries and hardens in 2–3 years. What was a paste-like layer in close contact when new gradually turns dry through thermal cycling (repeated heating and cooling).

Degradation tends to speed up in setups that run long, heavy loads such as AI processing or gaming.

Sometimes the application was poor to begin with

With prebuilt (BTO) PCs, or even a self-built PC where the initial application was sloppy, heat transfer can be inadequate from the very start. If you feel like “the temperature seems high, even from when I bought it," it’s worth a look.

Recommended paste comparison (5 products)

[kimono_product id="15850″]
[kimono_product id="15851″]

How to read this table:

• A higher thermal conductivity (W/mK) means higher theoretical heat-dissipation performance. Note that Noctua does not publish a thermal-conductivity figure for the NT-H1, so the value in the table is a reference figure cited in reviews and elsewhere. Judging by comparison testing in reviews, the actual temperature difference between products often comes to just 1–3°C.
• Viscosity directly affects how easy the paste is to apply. Softer ones spread easily and suit beginners; firmer ones can be applied thin and evenly but take some getting used to.
• Estimated lifespan is an average of manufacturer ratings and user reviews. It varies with your environment (temperature, how often you run heavy loads).
• Ease of application: ★5 is the easiest. If it’s your first time, choosing ★4 or above makes mistakes less likely.

Weighing up the review ratings, if you’re repasting for the first time, the Noctua NT-H1 (3.5g) looks easy to handle. It rates highly for ease of application, with a viscosity that’s just right and easy to spread. If you’re choosing on value, the Arctic MX-4 (4g) is a strong pick at about ¥800 with a generous volume. The go-to is the Thermal Grizzly Kryonaut (1g), high-performance at 12.5 W/mK. If you want both performance and long life, the Thermalright TFX (2g) pairs a high thermal conductivity of 14.3 W/mK with an estimated 5–7 year lifespan. When I repaste mine, since it’s my first time I’ll prioritize ease of handling and start with the NT-H1 or MX-4.

For reference, the Thermal Grizzly Hydronaut (1g, about ¥1,200) has a higher viscosity than the Kryonaut and resists running, so it’s said to suit those who want stability in a vertically mounted case or a high-temperature environment.

[Bar chart data] Thermal conductivity & price comparison

[Bar chart data] CPU thermal paste: thermal conductivity & price comparison
Chart type: bar chart (dual axis)
X axis: product / Y axis (left): thermal conductivity (W/mK) / Y axis (right): price (¥)

Product                       Thermal cond. (W/mK)  Price (¥)
Arctic MX-4                  8.5             800
Noctua NT-H1                 8.9             1000
TG Hydronaut                 11.8            1200
TG Kryonaut                  12.5            1500
Thermalright TFX             14.3            2000

Bar colors: thermal conductivity=blue, price=orange

How to read the graph:

• The blue bars (left axis) are thermal conductivity. You can see the trend of higher performance and higher price toward the right.
• The orange bars (right axis) are price. From the MX-4 (¥800) to the TFX (¥2,000) there’s a 2.5× price gap, but thermal conductivity ranges 8.5 → 14.3, only about a 1.7× difference.
• Value guide: compared by thermal conductivity per yen, the MX-4 (0.0106 W/mK per yen) is the best value, followed by the Hydronaut (0.0098) and then the NT-H1 (0.0089).
• Since the temperature difference between products in comparison reviews is said to be just 1–3°C, the practical conclusion seems to be “the cheap, easy-to-apply MX-4 or NT-H1 is plenty."

What you need for repasting & cost breakdown

Cost patterns summary (as of April 2026)

The minimum setup is for people who just want to repaste as cheaply as possible. The MX-4 and Kimwipes alone come to about ¥1,100. If the old paste is dried on and hard to remove, add anhydrous ethanol or a Sanwa Supply OA cleaner.

The recommended setup is the best-balanced combination for a first-timer. The NT-H1’s ease of application, plus reliable cleaning with anhydrous ethanol + Kimwipes, is said to make mistakes unlikely.

For wiping, the Kimwipe S-200 (about ¥300) is the standard. Tissues and kitchen paper can leave fibers on the CPU surface, but with Kimwipes there’s no worry about leftover fibers.

Application tips

From here, I’ve pulled together a general repasting procedure based on each manufacturer’s official instructions and commonly accepted methods.

1. Remove the old paste

Once the CPU cooler is off, wipe the old paste from the CPU surface and the cooler-side heatsink with a Kimwipe dampened with anhydrous ethanol. If it’s dried and hardened, use a bit more ethanol to soften it first, then wipe — it comes off cleanly.

2. Put a rice-grain-sized dot in the center

Squeeze out one rice grain’s worth of new paste onto the center of the CPU. That’s enough.

3. Lower the cooler straight down to spread it

Lower the CPU cooler straight down onto the paste and press it into even contact. There’s also a method of spreading it with a spatula, but placing a center dot and letting the cooler’s pressure spread it is said to be the least prone to uneven coverage.

Don’t over-apply

Reported temperature improvements

Here’s one example of reported temperature changes before and after repasting (report conditions: room temperature 25°C, Thermal Grizzly Kryonaut used). Looking at reviews and user reports, the improvement often lands in the 5–15°C range.

[Bar chart data] Before / After temperature comparison (reported example)

[Bar chart data] CPU temperature before and after repasting (reported example)
Chart type: bar chart (grouped)
X axis: CPU state / Y axis: CPU temperature (°C)

CPU state            Before (°C)  After (°C)
Idle                 45            38
Light load           55            47
LLM inference        82            73
Stress test          88            76

Bar colors: before=red, after=blue
Reference line: 90°C (rough onset of thermal throttling) shown as a red dotted line

How to read the graph:

• The red bars are CPU temperature before repasting, the blue bars after. In this reported example, the temperature drops clearly in every state.
• The difference is largest under heavy load. Against −7°C at idle, the stress test shows −12°C — the higher the load, the bigger the improvement. This is because the thermal resistance from degraded paste affects things more prominently in high-heat situations.
• The red dotted line (90°C) marks the rough threshold for thermal throttling. The pre-repaste stress test (88°C) is close to this line, and any further degradation would make the CPU automatically lower its clock, causing a performance drop.

When temperatures drop, the CPU cooler’s fan RPM can be held down too, so there are also many reports of “it got quieter." If the fan is noisy under heavy load, repasting alone may fix it.

When should you repaste? A decision checklist

The general guideline is once every 2–3 years, but the checklist below helps you judge “is it needed right now?"

Temperature-based criteria

Situation-based criteria

• The CPU cooler’s fan has become clearly louder than before (RPM is up)
• CPU temperature rose even though the case and room temperature haven’t changed
• It’s been 2 years or more since you bought a BTO PC without changing the paste
• 3 years or more have passed since you built a self-built PC
• You put long, heavy loads on it daily with AI processing, gaming, etc. (faster degradation)
• You removed and refitted the CPU cooler (repasting is mandatory when refitting)

If even one of the above applies, consider repasting. If any temperature-based item at ★★★ or above applies, prompt action is recommended.

For someone like me who puts long, heavy AI loads on it every day, it seems wise to consider repasting before the 2-year mark. Neither the cost nor the effort is large, so “look into it when it starts to bother you" is a fine mindset.

Summary

Repasting the CPU is an excellent-value bit of maintenance: ¥1,100–3,800 (as of April 2026), about 15 minutes of work, and many reports of CPU temperatures dropping 7–12°C.

Especially if you’ve never repasted a BTO PC since buying it, or it’s been 2 years or more since you built a self-built PC, it seems worth considering.

If you’re unsure which paste to pick, start by considering the Noctua NT-H1, which is about ¥1,000 and rates highly for ease of application.