PSU Sizing for Overclocking: Headroom, 80 Plus and Shutdown Clues

Why a failing stress test can be a power story

Prime95’s own documentation lists cooling and power among the things a Torture Test run exercises alongside the CPU and RAM, and the reasoning is plain. Small FFTs, the maximum power and heat preset, asks the processor to draw heavily and continuously, the voltage regulators on the motherboard convert that power under sustained load, and the power supply must deliver all of it, plus the graphics card, drives, fans and pumps, without its output sagging or its protection circuits deciding something is wrong. A supply that is ageing, undersized or simply optimistic in its labelling can coast through desktop work and gaming bursts for months, then meet a two-hour plateau of maximum draw and give up in the first hour. The failure signature matters. When a Prime95 worker stops with a rounding error, the machine computed wrongly and stayed alive to report it, which points at stability of settings, memory or cooling. When the whole machine cuts out instantly or reboots with nothing recorded, the machine did not stay alive to report anything, and power delivery belongs at the front of the suspect list. The wider crash investigation, including blue screens and their different meaning, is in Prime95 crashes and blue screens and the troubleshooting hub.

Sizing by whole-system estimate, with headroom as a principle

The reliable sizing method is arithmetic, not folklore. Add up the realistic maximum draw of the parts that can be busy together: the processor at its configured power limits rather than its headline figure where those differ, the graphics card, the motherboard and its regulators’ overhead, storage, fans, pumps and USB devices. Online estimator tools from power supply and case makers can help assemble that total, provided you feed them your actual parts and overclock intentions rather than a fantasy build. Then choose a quality unit rated comfortably above that total. The headroom is not waste; it serves four purposes at once. It absorbs transient spikes, the brief surges modern processors and graphics cards produce that averages never show. It keeps the supply working in the middle of its range rather than at its ragged edge, which is kinder to its components and usually quieter. It leaves room for the next upgrade, so a new graphics card does not force a second power supply purchase. And it covers ageing, because a supply’s comfortable output in year five is not quite its output when new. This guide names no specific wattage figure and recommends no specific models, because the right rating follows from your parts list and any table printed here would be wrong for most readers; the principle, estimate honestly and buy quality headroom above the estimate, is the portable part.

What 80 Plus does and does not tell you

Read the badge, then keep reading. A reputable maker’s unit with a long warranty, reviewed well by testers who actually load it, is a sounder purchase than an unknown unit wearing a higher efficiency tier. Efficiency decides how much extra power becomes waste heat inside the supply; it does not decide whether the supply survives your overclock.

Before you replace the supply, rule out the impostors

Shutdowns have other authors, and a careful tester checks them before spending. Overheating can shut a machine down protectively, so confirm temperatures and throttling behaviour first with the help of Prime95 temperatures explained. A marginal overclock or undervolt can also reset a machine under load, so retest at fully standard settings: a machine that only ever cuts out when tuned has accused its settings, not its supply. Loose or heat-damaged power connectors, a failing power strip, and a graphics card transient during mixed loads all belong on the list too. The supply becomes the prime suspect when shutdowns happen at standard settings, under CPU-only sustained load, with temperatures under control, and especially when the unit is old, small for the build, or of unknown origin. Substituting a known-good, adequately rated unit is then the decisive test, because it changes exactly one variable.

Buying a replacement, with the disclosure up front

Sponsored: this section contains a sponsored placeholder link, as explained on our affiliate disclosure page. Buy by the method above: your whole-system estimate, comfortable headroom, a reputable maker, evidence from load-testing reviews, and a warranty you would actually use. You can compare candidate units and current prices at Newegg via our sponsored placeholder link, which currently leads to the Newegg homepage while that affiliate programme awaits approval, and cross-check the same shortlist elsewhere before ordering. No model table appears here deliberately; a recommendation detached from your parts list would be theatre, not advice. After fitting the new unit, repeat the same Prime95 run that used to kill the machine, for at least as long as the run that failed, and keep the results file. A shutdown problem is fixed when the original test passes, not when a new box is on the desk. More hardware guidance for testers is collected in the Cooling and Hardware section.

Best DDR5 for Stability Testing: How to Choose Memory That Holds Up

Stability is a property of the combination, not the box

Memory does not work alone. Every DDR5 kit runs under the direction of a memory controller inside the processor, across traces on a particular motherboard, and the three parts succeed or fail together. A kit that is perfectly stable in one machine can be marginal in another whose controller is a weaker example of the same model, which is why copying a settings screenshot from a forum so often disappoints. When you choose memory with stability in mind, you are really choosing how much work to ask of that combination. Conservative rated speeds and relaxed timings ask little and usually succeed without drama. Aggressive ratings ask a great deal, and the payment comes due during exactly the long, memory-heavy testing this site is about. Neither choice is wrong; they are different priorities, and a stability-focused buyer should know which one they are making before the kit arrives.

The dimensions that decide the outcome

EXPO and XMP: stored overclocks, not factory promises

DDR5 kits carry profiles, called EXPO on AMD platforms and XMP on Intel platforms, that store the maker’s rated speed and timings for the board to apply in one step. Applying the profile is genuinely convenient, but it is important to understand what it is: the kit maker’s claim about the memory, applied to a controller and board the kit maker never saw. The stored profile raises memory speed beyond the processor’s base specification in most performance kits, and voltages and secondary timings move with it. Sometimes the result is immediately stable for years. Sometimes it fails only under sustained memory load, or only when all slots are warm, or only in a particular board. None of that makes profiles a bad feature; it makes them a starting point whose endpoint is validation. If a profile proves unstable, the standard, respectable fallbacks are to run the same kit at a lower speed, to relax timings, or to run the platform’s standard settings, each of which trades a little theoretical performance for a system you can trust.

How to verify the memory you chose

Prime95 contributes two presets to memory validation. Large FFTs use data sets too large for the processor caches, so they stress the memory controller and RAM continuously, and Blend, the default, mixes those large ranges with cache-sized work; the test menu itself notes that a Blend failure while Small FFTs passes points toward memory or its controller. Run the memory-focused sessions long, following the official 6 to 24 hour guidance for a thorough check, because memory faults are famously patient. Prime95 is not the whole answer, though: a dedicated bootable memory tester examines the RAM itself outside the operating system, and the division of labour between the two is set out in Prime95 versus MemTest86, while the detailed in-system method lives in the Prime95 memory test guide. A sensible order is to validate the bootable test first for faulty hardware, then validate in Prime95 for the whole memory path under real heat and load.

Buying, kept deliberately restrained

Sponsored: this section contains a sponsored placeholder link, explained on our affiliate disclosure page. Rather than pointing at specific kits with prices and performance claims this page cannot verify for your system, the honest buying advice is dimensional: pick capacity first, prefer two sticks for that capacity, choose a rated speed your platform can plausibly hold, favour kits on your motherboard’s qualification list, and keep enough budget flexibility to exchange a kit that fails validation. You can compare current DDR5 kits and prices at Newegg via our sponsored placeholder link, which currently leads to the Newegg homepage ahead of that affiliate programme’s approval, and check the same shortlist against other retailers before deciding. Whichever kit arrives, the purchase is not complete until it has passed the verification above at the settings you actually intend to run; a kit that passes only at standard settings has told you its real specification for your machine, and believing it is not a defeat but the test working as intended. More stability and hardware guidance is collected in the Cooling and Hardware section.

Undervolting Guide: Lower Heat Without Losing Stability

Why less voltage means less heat

A processor’s power draw rises with both clock speed and voltage, and voltage is the lever with the stronger effect: at the same frequency, a chip supplied with less voltage does the same arithmetic while converting less electrical energy into heat. That is the whole appeal. Factory voltage settings are chosen to suit every chip that leaves the line, including the weakest examples, so many individual processors can complete the same work on somewhat less. The difference shows up exactly where Prime95 users feel it: sustained loads run cooler, fans run slower, and processors that were brushing their thermal limit gain headroom without giving up speed. Undervolting is therefore often a better first move than underclocking for a hot system, because done correctly it trades nothing away. The phrase to hold onto is done correctly, because the margin being spent is the manufacturer’s safety margin, and it is spent invisibly until the moment it runs out.

Why an undervolt is not finished until Prime95 says so

An undervolted machine usually fails politely at first: it boots, it browses, it may even game, because light loads do not stress the marginal transistors the way a continuous maximum load does. Prime95’s Torture Test is built for precisely this gap. It runs known calculations and compares every result with the correct answer, so a voltage that is almost enough reveals itself as a rounding error or a stopped worker rather than waiting to corrupt something you care about. The official guidance in stress.txt recommends running the Torture Test for 6 to 24 hours for a thorough check, noting that unstable systems often fail within minutes while others take hours to expose a problem, and that short runs of 10 to 15 minutes suit exploring each tuning step before the long final test. That maps neatly onto undervolting practice: short runs while you step the voltage down, then one long, unbroken run at the setting you intend to keep. The duration reasoning is set out in full in how long to run Prime95, and the line between a genuine error and mere borderline behaviour is discussed in Prime95 errors versus an unstable system.

The risks, stated plainly

Two boundaries keep undervolting safe rather than adventurous. First, this guide deliberately gives no voltage figures and no model-specific values, because the amount any given chip tolerates is individual, and a number copied from another system is how careful tuning turns into someone else’s settings on your silicon. Second, know how to undo the change before you make it: note your original settings, change one value at a time in small steps, and be ready to restore defaults if the machine will not boot cleanly. Undervolting settings live in firmware setup screens or in your platform’s tuning utility, in sections concerned with processor voltage and frequency; the exact layout differs by maker and generation, so follow your own board or laptop maker’s documentation for where those controls sit rather than directions written for different hardware.

How undervolting relates to AVX and to cooling

AVX-heavy loads draw the most power, so they are where an undervolt runs out of margin first, and where platform features such as an AVX offset interact with your settings; that relationship is covered in Prime95 and the AVX offset. Undervolting also complements rather than replaces cooling work. Lower voltage reduces the heat produced; it does nothing for dried paste, a clogged filter or a case that cannot exhaust warm air, and a system that overheats at stock settings has a cooling fault to fix first, not a voltage project to start. Read the two subjects together in the Cooling and Hardware section, and sequence the work honestly: sound cooling first, then voltage tuning, then the long validation run that proves the combination.

A conservative workflow, start to finish

Record the stock settings and confirm the system passes Prime95 at them, so you have a clean baseline. Reduce voltage by one small step and run a short Small FFTs session to catch an immediately impossible setting. Repeat, stepping down gradually, and the first time any worker errors or the system misbehaves, step back up to the last clean setting. Then run the long validation, Blend for breadth or the preset matching your workload, for the full official window at that final setting, and keep the results file. If the long run fails, the setting was never finished, however many short runs it survived. That discipline is the entire difference between undervolting as engineering and undervolting as gambling.

Case Airflow for Stress Tests: Keep Prime95 Runs From Heat-Soaking

Why stress tests punish case airflow more than games do

A game loads the processor in waves and the graphics card in parallel, and sessions include menus, loading screens and quieter scenes. A Prime95 Torture Test run is a single, unbroken plateau of CPU heat, often for hours. In the first minutes, a good cooler holds temperatures easily because the air around it is still at room temperature. As the run continues, that air warms. The cooler now breathes pre-heated air, so its output temperature rises even though nothing about the cooler changed, and the warmer case air also bathes the memory, the motherboard voltage regulators and the drives. This slow drift is heat soak, and it is why a system can look comfortable in a ten-minute check and be throttling by the end of the official long-run window. Judging airflow on a short run is therefore a category error: airflow is precisely the part of the cooling chain whose quality only shows over time. If your temperatures climb steadily instead of levelling off, work through Prime95 overheating and 100C readings alongside this guide.

Intake, exhaust and a clear path between them

The working model is simple to state and easy to get wrong in detail. Cool air should enter, usually at the front and bottom of the case, pass across the components that need it, and leave, usually at the rear and top, carrying heat with it. Every fan should agree with that direction; a single reversed fan fights the stream and creates a warm eddy instead of flow. Obstructions matter more than fan count: a neat bundle of cables in the main chamber, a drive cage blocking the front intake, or a case pushed flush against a wall so its exhaust has nowhere to go can each undo an extra fan’s contribution. Dust filters and dust itself belong in the same audit, because a clogged intake filter quietly throttles the whole system. Before a long validation run, clean the filters, confirm every fan actually spins, and check that intake fans are not smothered by the surface the case stands on.

Fan curves for a load that does not fluctuate

Fan curves decide how aggressively case and cooler fans respond as temperatures rise, and stress testing exposes two common curve mistakes. The first is a curve tuned for silence at desktop loads, with fans barely turning until temperatures are already high, then surging; under a sustained load the system oscillates and soaks. The second is a curve tied to a temperature sensor that barely moves during a CPU test, leaving case fans idle while the processor cooks. The general approach that serves stress tests is a smooth ramp that reaches strong airflow well before the processor approaches its limit and then holds it steadily, because the load being cooled is steady too. Exact settings belong to your motherboard and your noise tolerance, and no universal curve is quoted here; what matters is the shape, a decisive ramp and a steady top end, verified by watching a long run rather than a quick spike. The meaning of the temperature readings you are responding to is covered in Prime95 temperatures explained.

When the answer really is better fans

Sometimes the audit ends honestly: the path is clear, the filters are clean, the curves are sensible, and the installed fans simply move little air quietly. That is the point at which replacement case fans earn their cost, particularly in the intake positions feeding the CPU cooler. A widely used premium option is the Noctua NF-A12x25, Noctua’s flagship 120mm fan, chosen by builders who want strong airflow with restrained noise character. Sponsored: the following link is an Amazon affiliate link and this site may earn a commission if you buy through it, at no extra cost to you; see our affiliate disclosure page. This site has no affiliate partnership with Noctua, ARCTIC, be quiet! or Thermalright, and none of them runs a public affiliate programme we can join, so the recommendation rests on the fan’s general positioning rather than any commercial arrangement: see the Noctua NF-A12x25 on Amazon. No airflow or noise measurements are quoted for it here, because honest figures depend on the case, the filter and the speed it runs at; judge candidates by your own constraints and retest after fitting. One caution first: replacement fans cannot rescue a case whose panels have almost no ventilation openings, and in that situation a case change, not a fan change, is the truthful recommendation.

Verify airflow the way you verify stability

Change one thing at a time, then rerun the same Prime95 preset for long enough for heat soak to declare itself, and compare where temperatures level off. A change that only improves the first five minutes has improved the wrong thing. Record the levelling point, the fan behaviour and the room conditions alongside the run, so the next change builds on evidence. Cooler-side improvements are covered in CPU coolers for stress testing, and more cooling work is collected in the Cooling and Hardware section.

Prime95 Temperatures Explained: What the Numbers Actually Mean

Why there is no single safe number to quote

It would be convenient if one temperature figure separated safe from unsafe for every processor, and many pages online pretend one exists. Prime95’s documentation does not provide such a figure, and neither Intel nor AMD applies one number across their ranges: limits are set per model, because silicon, packaging and intended use differ. A compact laptop processor, a desktop flagship and a workstation part can each have a different specified maximum. Quoting a generic threshold therefore misleads in both directions, alarming owners of models designed to run near it and reassuring owners of models whose real limit is lower. The correct habit is slower but honest: find your exact processor model, open the manufacturer’s specification page for it, and note the stated TjMax or maximum temperature. That figure, and only that figure, is the ceiling your readings should be compared against. The broader safety question, including what a long run does and does not risk, is treated in Is Prime95 safe?

Why Small FFTs produces the highest temperatures

The Torture Test presets differ by where their data lives. Smallest FFTs work inside the L1 and L2 caches with high power, heat and CPU stress. Small FFTs cover the L1, L2 and L3 caches and are described by the program itself as the maximum power, heat and CPU stress preset: the data stays on the processor, the cores never wait on memory, and arithmetic runs continuously at full rate. Large FFTs use data too large for the caches, so the processor spends part of its time fetching from RAM, shifting the burden toward the memory controller and memory. Blend, the default, mixes all of the above. This is why the same machine, at the same settings, shows its highest CPU temperatures in Small FFTs and lower ones in Large FFTs, and why Small FFTs is the right preset when the question is specifically about cooling headroom. It is also why a temperature complaint is meaningless without the preset attached: always record which preset produced the reading.

The AVX context, labelled as the second-hand claim it is

Modern Prime95 versions use AVX instructions where the processor supports them, and AVX code paths raise power draw under continuous load. A long-running community source, the Tom’s Hardware Intel Temperature Guide, reported that on older Intel Core processors, from the 2nd to the 8th generations, versions after 26.6 running AVX code could produce temperatures up to about 20 degrees Celsius higher than the older pre-AVX version 26.6, and called those temperatures unrealistic compared with typical workloads. Two labels belong on that figure whenever it is quoted, including here: it is a second-hand community guide figure, not an official GIMPS or Intel specification, and it describes older platforms. Modern processors manage AVX loads with offset and power-limit features that change the picture, so the figure must not be carried across to a current machine as a prediction. What survives from it is a modest, useful point: an AVX Prime95 run is an extreme load by design, and a temperature that looks startling next to a game can still be a normal result of that design. Deciding whether yours is acceptable still comes back to your model’s specified limit.

What to monitor during a run

Most builders watch these values in HWiNFO, a free monitoring tool available from hwinfo.com. HWiNFO is named here purely as a practical companion for reading sensors; this site has no partnership with it and earns nothing from the recommendation. Whatever tool you use, watch behaviour over time rather than a single peak: a temperature that rises and then levels off describes a cooling system that has found its balance, while one that climbs steadily for as long as you watch describes a system that has not, and that pattern matters more than any single snapshot. Old guidance in Prime95’s stress.txt also carried a period note about first-generation Ryzen parts and heat-related instability; it belongs to its 2019 context and is not a limit for current processors, which is one more reason per-model specifications are the only sound reference.

Putting a reading into a decision

A temperature reading becomes useful when it leads to the right action. Below your model’s limit, with no throttling and no worker errors, a hot Small FFTs run is the test doing its job, and the machine is behaving as designed under an extreme load. At the limit with throttling, the cooling chain needs work before long runs continue: mounting, paste, airflow and cooler capacity, in that order of cost, are the subjects of Prime95 overheating and 100C readings and the wider Cooling and Hardware section. A worker error at any temperature is a stability problem first and a temperature problem second; lowering heat may help, but the error, not the thermometer, is the finding to chase.

Frequently asked questions

Is there an official safe Prime95 temperature?

No. Prime95’s documentation publishes no universal safe temperature that applies to every processor. The limit that matters is the one your CPU maker specifies for your exact model, often listed as TjMax or a maximum operating temperature on the manufacturer’s specification page.

Why does Prime95 Small FFTs run hotter than other presets?

Small FFTs is the preset the program itself describes as producing maximum power, heat and CPU stress across the L1, L2 and L3 caches. Data stays in cache, the cores work continuously, and power draw and heat are at their highest, so it is the preset that reveals a CPU temperature ceiling.

What should I watch while Prime95 runs?

Watch the CPU package temperature, whether clock speeds are being reduced under load (thermal throttling), and whether any worker stops with an error. A monitoring tool such as HWiNFO shows these values while the test runs.

My temperatures are much higher in Prime95 than in games. Is something wrong?

Not necessarily. Prime95, especially Small FFTs, is a deliberately extreme and continuous load that most real programs do not match. Community guides have long noted higher temperatures under AVX Prime95 loads on older platforms, but the useful comparison is against your own CPU maker’s limit for your model, not against a game.

Thermal Paste Guide: When to Replace It and How to Apply It

What the paste is actually doing

Two pieces of metal that look flat are not flat at the scale heat cares about. Under magnification, both the processor heat spreader and the cooler base are landscapes of tiny ridges and valleys, and where they meet, much of the contact is actually trapped air. Thermal paste is a heat-conducting compound designed to fill those gaps so heat crosses from the spreader into the cooler instead of stalling in air pockets. It is not glue, it is not meant to be a thick cushion, and more is not better: the metal surfaces should sit as close together as the mounting allows, with paste occupying only the imperfections. That is why the guiding principle for application is thin and even, and why mounting pressure, applied evenly, matters as much as the paste itself. A premium cooler mounted over a thick, uneven paste layer can perform like a much cheaper part.

Signs it is time to replace the paste

The clearest sign is a slow story rather than a sudden failure: temperatures under the same Prime95 load are higher than they used to be on the same settings, fans work harder to hold the same point, and nothing else in the system has changed. Paste ages through heat cycling; over years it can dry, harden or migrate away from the centre of the spreader, a process often called pump-out. The second sign is mechanical: any time you remove the cooler for any reason, the old paste layer is disturbed and should be cleaned off and replaced rather than reused, because reseating on a broken, partially dried layer traps air exactly where you least want it. The third sign is visible during that removal: paste that is cracked, chalky or mostly missing from the middle of the spreader has already told you it was part of the problem. If your readings are alarming right now, read Prime95 overheating and 100C readings first, then repaste as one step in that wider check rather than as a blind ritual.

How to apply it: thin, even, complete

Different builders swear by different application patterns, and the honest truth is that the principle matters more than the pattern: thin, even and complete coverage of the heat spreader, with the cooler pressing it into place, is what every sound method is trying to achieve. What goes wrong is usually one of three things: far too much paste, which squeezes out and insulates at the edges; far too little, leaving dry corners of the spreader uncovered; or a rushed mount that tightens one side fully before the other, trapping a wedge of air. If you see squeeze-out after mounting, you used more than the job needed; clean up the excess so it cannot migrate onto the socket or board. There is no universal measured quantity worth quoting here, because spreader sizes differ between processor families, which is exactly why the thin-and-even principle is the reliable instruction.

How paste fits the bigger cooling picture

Paste is one link in a chain: heat spreader, paste, cooler base, heat pipes or cold plate, fins, fan air, case air, room air. Repasting fixes only its own link. If temperatures stay at the limit after a careful repaste, the limit lives elsewhere in that chain, and the next checks are cooler capacity and case airflow. The cooler side is covered in CPU coolers for stress testing, and what the temperature readouts themselves mean, including why Prime95 publishes no universal safe number, is covered in Prime95 temperatures explained.

Buying paste, with the disclosure up front

Sponsored: the link in this section is an Amazon affiliate link, and this site may earn a commission if you buy through it, at no extra cost to you; see our affiliate disclosure page for how affiliate links work here. This site has no affiliate partnership with ARCTIC, Noctua, Thermalright or be quiet!, and none of them runs a public affiliate programme we can join. A widely used, general-purpose option is ARCTIC MX-6 thermal paste on Amazon; MX-6 is ARCTIC’s current mainstream thermal compound, and it is linked here as a straightforward example rather than on the basis of any test figures, which this page does not quote. Any reputable modern paste from a known maker, applied thinly and evenly under a well-mounted cooler, serves a stress tester well. Paste is the cheapest part of the cooling chain, so there is little reason to economise with an unknown compound of unclear composition.

The repaste checklist before your next long run

Clean both surfaces fully, apply a small amount, mount evenly, then verify under a sustained Prime95 run while watching temperatures level off. Keep a note of the date you repasted; when temperatures creep again years later, that note turns guesswork into arithmetic. More guides like this one live in the Cooling and Hardware section.

Best CPU Coolers for Stress Testing: Air Cooling That Survives Prime95

Why Prime95 is the stress test that exposes coolers

Most everyday loads let a processor breathe. Games spike and settle, office work idles between bursts, and even rendering finishes and lets the cooler recover. The Prime95 Torture Test does none of that. It keeps every worker calculating continuously and compares each result against a known correct answer, so the load is both heavy and unbroken. Within that test, the presets are not equal. Smallest FFTs keep data in the L1 and L2 caches with high power and heat. Small FFTs extend that across the L1, L2 and L3 caches and are officially the maximum power, heat and CPU stress preset. Large FFTs shift pressure toward the memory controller and RAM, and Blend, the default, mixes all of it. A cooler that looks adequate in a short benchmark can slowly heat-soak during a long Small FFTs run until temperatures level off at a point you do not like, or the processor starts reducing clock speed to protect itself. That is why cooler advice written for gaming loads often fails Prime95 users, and why this guide treats a sustained torture run as the design case. If your current setup already hits its limit within minutes, work through Prime95 overheating and 100C readings before you spend anything, because mounting and airflow faults imitate an undersized cooler.

The four buying dimensions that actually decide the purchase

Work through that table in order. Fit is a hard filter: measure, or look up both stated figures, before falling for any particular model. Memory height is the filter buyers most often forget, especially with tall RGB memory kits. Noise is a genuine preference rather than a flaw; a test bench in a workshop and a PC beside a desk call for different choices. Budget comes last because the first three usually narrow the field to a handful of sensible candidates anyway.

Two strong air cooling options, described honestly

Noctua NH-D15 is Noctua’s flagship dual-tower air cooler, a long-running reference point in this category, built around two large fin stacks and a focus on quiet operation for its size. It suits builders who want maximum air cooling headroom and are willing to pay flagship-tier pricing, provided their case and memory clearances allow it. Thermalright Peerless Assassin 120 is Thermalright’s high-value dual-tower air cooler, widely chosen because it brings the same dual-tower layout to a much lower budget tier. It suits builders who want stress-test-capable cooling without flagship spending. Neither description comes with a temperature figure, a noise figure or a ranking between them, because honest numbers depend on your processor, your case and your fan settings; any page that promises one universal result for either cooler is inventing precision. Choose between them by the dimensions table above: clearance, memory, noise preference and budget.

Do not skip the paste and the mounting

A cooler upgrade installed over old, dried-out thermal paste with uneven mounting pressure wastes much of its advantage. Thermal paste fills the microscopic gaps between the processor heat spreader and the cooler base, and it ages. If you are changing coolers anyway, clean both surfaces and apply fresh paste as part of the same job; our thermal paste guide covers when paste needs replacing and how to apply a thin, even layer. Tighten mounting hardware gradually and evenly, following the cooler maker’s instructions, so pressure spreads across the heat spreader rather than tilting the cold plate. Then retest with the same Prime95 preset you used before the change, so the comparison means something.

Where to buy, with the disclosure up front

Sponsored: the product links in this section are Amazon affiliate links. This site may earn a commission if you buy through them, at no extra cost to you, and how that works is explained on our affiliate disclosure page. Just as plainly: this site has no affiliate partnership with Noctua, Thermalright, ARCTIC or be quiet!. None of those brands runs a public affiliate programme we can join, so their products are recommended here on positioning and reputation alone, and purchases are made through Amazon. See the Noctua NH-D15 on Amazon, the Thermalright Peerless Assassin 120 on Amazon and ARCTIC MX-6 thermal paste on Amazon. Prices and stock change constantly, so this page quotes neither; check the product page for the current figures. For a second price reference, you can also compare coolers at Newegg via our sponsored placeholder link, which currently leads to the Newegg homepage and will switch to a tracked link once that programme is approved.

After the upgrade, prove it the same way it failed

Run Small FFTs again and watch the run for longer than feels necessary. Temperatures should level off rather than climb steadily, and no worker should stop with an error. If readings are still at the processor’s limit, the bottleneck may have moved: case airflow that cannot feed the cooler is the usual next suspect, covered in case airflow for stress tests, and the meaning of the temperature numbers themselves is covered in Prime95 temperatures explained. More cooling and hardware guides are collected in the Cooling and Hardware section.