How to Read CPU Cooler Specs: Four Numbers That Decide Your Purchase

Buying the wrong cooler costs more than buying the wrong mouse. A mouse that does not suit you gives you a sore wrist for a few days. A cooler chosen from the wrong number gives you fans pinned at full speed, or a CPU that drops out of PL2 under load and quietly gives you frames you paid for. Meanwhile almost every number on the spec sheet — TDP, dB(A), mmH2O, RPM — is routinely misread.

This is not a temperature test. I do not own these four coolers and I do not have a rig that can load a 285K, so I am limiting this to what you can verify yourself on a manufacturer's page: what each number measures, which numbers must never be compared directly, and how to turn that into a decision for your own build. I opened all four ASINs on amazon.com/dp/ one by one while writing this (checked 2026-10-21). Where a listing does not state a number, I write "not listed" rather than filling it in. Amazon links here are affiliate links: you pay the same price, I earn a small commission.

⏳ TL;DR — Too Long? Read This Instead

Buy on "how much heat must this actually remove", not on the wattage printed in the spec sheet. The four product pages are Noctua NH-D15 G2, ARCTIC Freezer 36 A-RGB, be quiet! Pure Rock Pro 3 Black and Noctua NF-A12x25 G2 PWM, all ASINs re-checked on the product pages. Prices and stock depend on when you order, so this article quotes no price I did not verify.

Pushing past 200W and happy with a 168mm tower: the NH-D15 G2, an 8-heatpipe dual tower with two NF-A14x25r G2 fans and 2.56 mm H2O of static pressure, the highest figure among these four. The cost is RAM clearance: only 32mm with both fans installed.

Wanting a cooler you can compare on a like-for-like noise basis: the ARCTIC Freezer 36 A-RGB, whose official spec sheet publishes two wattage ratings, PI and NNPI, so the spec sheet itself answers "how much cooling at equal noise".

Want a noise curve you can verify yourself: the be quiet! Pure Rock Pro 3 Black, whose official datasheet gives 14.0 / 26.5 / 34.8 dB(A) at 50% / 75% / 100% PWM. Among these four it is the only one publishing a multi-point curve.

Replacing a fan on an existing tower or radiator: the NF-A12x25 G2 PWM. Note it is a fan, not a cooler. You still need a heatsink.

Layer 1: TDP is the CPU's baseline, not your cooling requirement

Intel's own definition of TDP carries an unusually blunt caveat. From the 12th generation onward, the TDP label was replaced with Processor Base Power, and under turbo or certain workloads such as AVX, the chip "can exceed the maximum TDP but only for a limited time, or Until the processor hits a thermal throttle temperature, or Until the processor hits a power delivery limit." In selection terms: TDP is the steady-state design baseline, not the peak your machine will draw for five minutes.

The harder question is how long that "limited time" actually is. The General Notes section of Intel's 13th-generation Raptor Lake datasheet states that the hardware default for PL1 Tau is 1 second, while the same document recommends a value of 28 seconds. So a CPU boosting briefly to PL2 has roughly a one-second window at PL2 under default firmware before it is pulled back to PL1. If you pick a cooler against a 250W transient peak, you may be paying for a one-second window while everything else runs at PL1.

One real datapoint makes this concrete. On Intel's ARK page, the Core Ultra 9 285T lists a Processor Base Power of 35W and, on the same page, a Maximum Turbo Power of 112W — two official numbers on one chip that differ by 3.2x. Looking at either one alone gives you the wrong headroom.

On AMD's side the comparable concept is PPT (Package Power Tracking), which caps package power and is not the same thing as TDP either. The wattage a cooler maker puts on a spec sheet usually tracks sustained PPT or PL1 behaviour more closely than the number printed on the retail box.

Layer 2: Normalise the noise basis before you compare dB(A)

This is the point I most want you to take away. Noctua's own FAQ contains what is essentially a consumer-facing controlled comparison. The standard NF-F12 PWM emits 22.6 dB(A) at its 1500rpm maximum, while the NF-F12 industrialPPC-2000 PWM emits 29.7 dB(A) at its 2000rpm maximum. But the official text immediately adds the qualifier that matters: "When both fans are adjusted to the same speed via PWM or voltage control though, the noise level is similar."

That directly undermines the obvious comparison. Comparing two coolers by their quoted maximum-speed noise measures the difference between their fans' top-end speeds, not the difference you will actually hear. And in real use the fan is almost never at maximum, because the motherboard or fan controller curve holds it somewhere around 30% to 60% during light load.

With that in place, here are the four noise figures in this article:

• The Noctua NH-D15 G2 official specifications give 24.8 dB(A) and 2.56 mm H2O, with a speed range of 0–1500 rpm.
• The Noctua NF-A12x25 G2 PWM listing gives 22.5 dB(A) at a maximum of 1800 rpm, a range of 360–1800 rpm, and a bundled Low-Noise Adaptor that lowers the ceiling from 1800 to 1500 rpm.
• The ARCTIC Freezer 36 A-RGB listing gives two P12 PWM PST A-RGB fans at 200–2000 rpm, stopping below 5% PWM. The official spec sheet does not list dB(A) separately in its main table.
• The be quiet! Pure Rock Pro 3 Black official datasheet, marked V2/25, gives 14.0 / 26.5 / 34.8 dB(A) at 50% / 75% / 100% PWM respectively, at 2000 rpm when fully loaded.

Put all four side by side and the conclusion "34.8 is louder than 24.8" proves nothing — not that the Pure Rock Pro 3 is noisier, not that the NH-D15 G2 is quieter. They are not even measured at the same operating point. The 34.8 is two fans at full speed; the 24.8 corresponds to a published ceiling of 1500 rpm. What actually compares is data like be quiet!'s, where one machine's curve is published across several duty cycles.

ARCTIC took a different approach and went further than anyone else. Its official FAQ states plainly that "As the TDP of the CPUs is only of limited significance for the expected cooling performance", then defines two indicators. PI (Performance Indicator) is the cooling capacity required at full speed to keep the heatspreader no more than 30 °C above room temperature. NNPI (Noise-normalised Performance Indicator) limits the fan speed so that noise lands at 28 dB(A) measured at 50 cm. The Freezer 36 A-RGB official page publishes exactly this pair as PI | NNPI: 245 | 219. Those two numbers can be compared across brands, because the noise level is already pinned to one fixed point.

Layer 3: mmH2O and CFM are different jobs

The other easily confused pair is airflow versus static pressure. Airflow, quoted in CFM or m3/h, describes how much air a fan moves in open space. Static pressure, in mm H2O, describes how much resistance it can push against. Dense fin stacks on tower coolers, radiator cores and dust filters are all high-resistance applications, and there static pressure is the number that decides whether you can hold temperature.

The official figures here: each of the NH-D15 G2's two NF-A14x25r G2 fans gives 155.6 m3/h (91.58 CFM) at 2.56 mm H2O; the be quiet! Pure Wings 3 gives 59.6 CFM / 101.2 m3/h at 2.41 mm H2O at 100% PWM; and ARCTIC markets the P12 PWM PST A-RGB as "pressure-optimised" on the product page without listing a static pressure figure in its main table.

Note that the NH-D15 G2 uses 140mm round-frame fans while the Pure Rock Pro 3 uses two 120mm units. Different fan count, different diameter, different speed ceilings — yet static pressure differs by only 0.15 mmH2O. Against a tower fin stack, one large slow fan and two small fast fans can land in nearly the same place thermally, even though their prices, noise curves and RAM interference have nothing to do with each other.

Layer 4: Size and clearance, check last but always check

The spec item most often omitted and most often regretted is how the cooler physically interacts with your case, RAM and VRM heatsinks. Official dimensions for these four: the NH-D15 G2 is 168mm tall with fans, 150mm wide, 152mm deep, and weighs 1525 g including fans; the Freezer 36 A-RGB is 104 × 126 × 159mm at 917 g; the Pure Rock Pro 3 is 139 × 124 × 155mm at 1077 g.

RAM clearance deserves its own line. The NH-D15 G2 listing states it plainly: the offset design gives 59mm of RAM clearance, but that is in single fan mode. With both fans fitted, only 32mm remains. If your memory has heat spreaders taller than 40mm, think twice about the two-fan option. be quiet! takes a different route: its listing highlights an adjustable front-fan height, using the offset layout to buy RAM and VRM compatibility.

One more comparison dimension is easy to overlook and that is warranty. Noctua and ARCTIC both carry 6 years on these models; the be quiet! Pure Rock Pro 3 is 3 years. A cooler is the sort of part you fit once and never replace, so that gap is a real long-term ownership cost.

The four, one by one

Noctua NH-D15 G2 — 8-heatpipe dual tower, rated by the manufacturer for 250W and beyond

Product page: Noctua NH-D15 G2, 168mm, dual 140mm round-frame fans, 6-year official warranty.

Real strengths: the highest static pressure of these four at an official 2.56 mm H2O; eight heatpipes, with the official product page noting that the extra thermal capacity from the extra heatpipes pays off most at extreme loads "250W and beyond"; the listing states NM-ISW1 spacers are included, which are used on LGA1700 to counteract CPU warpage; and the offset design clears the top PCIe x16 slot.

Real drawbacks: at 168mm it is the tallest tower here, so measure your case airflow clearance first. RAM clearance drops to 32mm with both fans fitted, which is the mistake most likely to surface after delivery. The SecuFirm2+ mounting system uses Torx, so when you move to a different platform, check that the bundled screwdriver still covers it.

ARCTIC Freezer 36 A-RGB — the one that bakes noise normalisation into the spec sheet

Product page: ARCTIC Freezer 36 A-RGB, official PI | NNPI = 245 | 219, 6-year warranty, 0.8 g of MX-6 thermal compound pre-applied.

Real strengths: the official spec sheet covers four sockets, LGA1851, LGA1700, AM5 and AM4. The Contact Frame screws into the backplate rather than loading the CPU, and the official text says this massively reduces mechanical stress on the chip. The side-opened fin stack lets the rear fan draw cool air from outside the case. The listing also states the cooler accepts most 120mm fans, giving you a wider replacement choice than a dual-tower design.

Real drawbacks: official and retailer figures disagree in a way worth understanding. The official ARCTIC page for the Freezer 36 A-RGB states PI | NNPI: 245 | 219, while retailer pages for the Freezer 36 Black, the non-RGB version, list PI 250 and NNPI 234. Those are different machines and different measurements, and mixing them produces wrong conclusions. The A-RGB version needs a 5V 3-Pin A-RGB header; older motherboards with only 12V 4-Pin RGB will simply not light up. And with a single tower and four heatpipes, the official 245W figure is quoted at full-speed noise, so in a noise-sensitive build read the NNPI figure of 219 instead.

be quiet! Pure Rock Pro 3 Black — the only one of the four publishing a three-point noise curve

Product page: be quiet! Pure Rock Pro 3 Black, 250W, dual tower with 6 heatpipes, 2 × 120mm PWM, 3-year warranty.

Real strengths: the official datasheet splits noise into 14.0 / 26.5 / 34.8 dB(A) at 50% / 75% / 100% PWM, which among these four is the only one that lets you plot a curve instead of reading a single endpoint. It is officially marked compatible with liquid metal compounds. The adjustable-height front fan in an offset layout is real: buyer comments on the listing mention a cooler that did not fit until the fan was moved, which is the interference problem solved. A Y-Cable for two headers is included.

Real drawbacks: 34.8 dB(A) is the highest full-speed figure of these four, and two fans in a dual tower at full speed is exactly what that number describes. Three years of warranty is the shortest here. The official "250W TDP" label uses the TDP framing, while Intel renamed TDP to Processor Base Power from the 12th generation onward, so the two numbering systems do not line up directly. The fan cable is 120mm, shorter than Noctua's 220mm, so cable management may need attention.

Noctua NF-A12x25 G2 PWM — this one is a fan, not a cooler

Product page: Noctua NF-A12x25 G2 PWM, 120 × 120 × 25mm, 360–1800 rpm, stops at 0% PWM.

Real strengths: the official press release dates this second-generation 120mm flagship to June 2025, with a Sterrox LCP impeller and a 0.5mm tip clearance. The bundled Low-Noise Adaptor caps maximum speed at 1500 rpm instead of 1800, which is an official, executable version of the "slow it down to quiet it down" strategy. The listing confirms it stops at 0% PWM, enabling semi-passive builds. The family also includes an LS-PWM low-speed variant, which the official release rates at up to 1100 rpm, and an Sx2-PP dual set whose two fans differ by roughly 50 rpm to suppress periodic humming when mounted side by side.

Real drawbacks: there is no heatsink, so it must be paired with an existing tower cooler or radiator and cannot be fitted on its own. The 22.5 dB(A) figure is the maximum-speed value from the listing, not the value at 1500 rpm, so do not compare it against your build with the Low-Noise Adaptor fitted. The official release lists both a standard PWM at 1800 rpm and an LS-PWM at 1100 rpm, and their noise behaviour differs substantially; the listing title does not always make it obvious which variant you are buying, so picking wrong is plausible.

FAQ

Q: Is a cooler rated well above my CPU's stated TDP guaranteed to be enough?

A: Not guaranteed, but not necessarily insufficient either. TDP is the steady-state baseline, and your CPU will exceed it under turbo or AVX workloads. How long depends on Intel's PL1 Tau (hardware default 1 second, official recommendation 28 seconds) or the AMD PPT setting. The more reliable approach is to use a vendor's cooling-capacity metric rather than a wattage label, such as ARCTIC's noise-normalised NNPI.

Q: So is 24.8 dB(A) or 34.8 dB(A) actually louder?

A: Those two numbers alone cannot be compared. The 24.8 dB(A) may correspond to a 140mm fan at a 1500 rpm ceiling, while the 34.8 dB(A) is two 120mm fans at full 2000 rpm. The measurement points differ. What compares is one cooler's curve across several PWM duty cycles, or a vendor's noise-normalised figure.

Q: My CPU temperature keeps climbing. Do I need to switch to water cooling?

A: First determine whether you are hitting a power limit or a cooling limit. If temperature rises while the CPU throttles and PL1/PL2 or PPT is constrained in the BIOS, a bigger cooler buys you little, because the bottleneck is the power ceiling rather than the heatsink. If the CPU sits at PL1 without downclocking, cooling capacity may genuinely be the limit. Monitoring software separates these two cases by showing whether you are pegged against a power limit.

Q: What does 32mm of RAM clearance on a dual-tower cooler mean?

A: It is the available height between the inner edge of the cooler and the top of the memory slots with both fans installed. If your memory heat spreaders are taller than that, either switch to single-fan mode (the NH-D15 G2 gives 59mm that way) or choose a model with an adjustable front fan, such as the Pure Rock Pro 3. Measuring your memory height before ordering beats swapping RAM after delivery.

Q: For a quiet build, should I simply pick the lowest quoted noise figure?

A: First confirm whether that figure is a full-speed value. If one product is quoted at full speed and another at half speed, a direct comparison misleads you. What actually shapes daily noise is the fan curve you set in the BIOS. Setting a slightly lower temperature ceiling and moving the curve's knee earlier usually does more than switching to a cooler quoted 2 dB lower.

Troubleshooting: the numbers look right but temperatures do not

Case 1: I picked the cooler using NNPI, and the machine still hits 95 °C and downclocks under load.

Check PL1 Tau first. If the time constant visible in your BIOS or Intel XTU is still the 1 second hardware default, the high-power window is very short. If someone has set it to 28 seconds or longer, the CPU spends most of its time at PL2, and the cooler's NNPI rating — measured at reduced speed for noise reasons — no longer describes the real load. Check the BIOS power limits before you check the cooler.

Case 2: after fitting a large tower, the RAM does not seat or the board does not boot with it.

This is a clearance issue, not a compatibility issue. First confirm whether it fits at all with both fans (the NH-D15 G2 is 32mm). If it fits, reseat the fans properly and retry; if it does not, drop back to single-fan mode. In other cases the cooler is fouling the VRM heatsink or the M.2 heatsink, in which case any bundled spacer heights need removing before assembly.

Case 3: the case is loud, yet each fan measures quiet on its own.

A common cause in multi-fan setups is two fans running at identical speeds, producing a beat frequency whose combined tone is harsher than either fan alone. Noctua addresses exactly this in its official release: the Sx2-PP version avoids periodic humming by offsetting the two fans by roughly 50 rpm. Another cause is structural resonance — if the frequency sits near the natural frequency of a case panel or drive, offsetting the speed ranges usually works better than adding rubber pads.

Wrapping up

TDP is the CPU's steady-state baseline, not your cooler's requirement. dB(A) needs a stated speed before it means anything. mmH2O decides whether a dense fin stack gets cooled. Size and RAM clearance are the last thing to check and the easiest thing to get wrong. Of the four coolers here, ARCTIC's PI and NNPI are the most useful habit to copy, because they pin noise to one point before comparing performance. be quiet!'s three-point curve is the most transparent practice. Noctua's LNA and LS-PWM are the most practical noise strategy: rather than chasing a quieter headline number, run the fan slower most of the time.

If you are still shopping in the order "check TDP, then check price", start with the site's guide to PC case selection to confirm airflow and maximum cooler height, then come back and run these four layers over the cooler itself.

👉 Join Xiaomi MiMo Platform: Leading AI model platform with cost-effective inference

👉 Join Aliyun AI: Top AI products with exclusive coupons for business innovation

📌 This article was AI-assisted generated and human-reviewed | TechPassive — An AI-driven content testing site focused on real tool reviews