How to Use Prime95 to Stress Test Your CPU

Prime95 is one of the most demanding CPU stress tests ever written, which is exactly why overclockers, system builders and repair shops still use it. This guide shows you how to run a Prime95 stress test on Windows from start to finish: what to prepare, which options to choose in the torture test window, how long to run it, how to tell a pass from a fail, and what to do when something goes wrong.

Before you start

A stress test is a deliberate worst case, so preparation matters more than the test itself. Spend five minutes on the following and you’ll avoid most of the confusion that new users run into.

• Get the right file. Download the official zip from the download page and extract it to a folder such as C:\Tools\Prime95.
• Install a monitor. HWiNFO (sensors-only mode) or Core Temp shows per-core temperature, package temperature, clock speeds and throttling flags. Prime95 itself doesn’t show heat.
• Close everything else. Games, browsers with dozens of tabs and background renders all muddy the result and add heat that isn’t part of the test.
• Save your work. If your system is unstable, a stress test is how you’ll find out, and an unstable machine can freeze or restart.
• Use AC power on a laptop and make sure the vents are clear.
• Know your limit. Look up your CPU’s maximum temperature (TjMax) on Intel ARK or AMD’s product page. Safe Prime95 temperatures explains how to use it.

Step-by-step: run a Prime95 stress test

1. Launch Prime95. Double-click prime95.exe. If SmartScreen warns about an unrecognised app, that’s normal for small utilities. Run it only if the file came from the official source.
2. Choose Just Stress Testing. The first window asks about joining GIMPS. Joining starts a long-running background workload, which isn’t what you want for a stability check. See the difference.
3. Open the torture test window. It appears automatically on first run. Later you can reach it from Options, Torture Test.
4. Leave the thread count at its default. The default stresses all of your logical processors. That’s the right choice for a worst-case run.
5. Pick a test type. Blend is the default and the best all-rounder. Choose Small FFTs when you want to focus on CPU cores and heat. The next sections explain why.
6. Press OK. Prime95 starts one worker per thread and the CPU load jumps to 100%.
7. Watch your monitor for the first five minutes. If temperatures climb toward your CPU’s limit, stop the test with Test, Stop.
8. Let it run for your chosen time, checking now and then. When you’re finished, stop it from the Test menu.

The torture test window explained

The dialog looks intimidating, but only a few settings matter for a normal stability test. Labels vary slightly between versions, so treat this as a guide to the concepts.

A note on that memory setting: according to PlayTool’s long-running Prime95 guide, if your hard disk light is on most of the time during Blend, it has allocated more memory than your PC physically has. Reducing the memory value in a custom test until the disk activity stops fixes it.

Which test type should you choose?

The choice comes down to where the data lives while Prime95 works. Small FFTs fit inside the CPU cache, so memory is barely touched and the cores do nearly all the work. Large FFTs spill out into main memory. Blend mixes both. That’s the one idea you need.

Newer versions also include a “Smallest FFTs” option that uses an even smaller footprint and is the most extreme for heat and power. Treat it with extra caution. The full comparison, including real measurements from published tests, is in Small FFTs vs Large FFTs vs Blend.

How long should you run Prime95?

There’s no official required time, and published advice varies widely. That’s why it helps to know what each length is good for.

Community threads on HardForum and Overclockers Club describe a common routine of about 8 hours of Small FFTs followed by about 24 hours of Blend. PlayTool suggests a few hours, or 24 hours to be thorough. Others point out that if you never load the PC for more than a couple of hours at a time, shorter tests are enough. These are opinions, not official guidance, and many of the threads are years old.

The honest trade-off is risk versus confidence. A gaming PC that crashes once a week is annoying. A workstation that silently corrupts a file is worse. Match the test length to the cost of a failure.

Temperatures and when to stop

Prime95 produces more heat than games, and that’s expected. What matters is whether your CPU stays under its limit without throttling. To put the heat in context, SkatterBencher measured a stock Core i7-13700K running Prime95 Small FFTs at about 253 W and an average of 90°C, using a 360 mm custom water loop (source). A smaller cooler would run hotter.

If you reach the limit quickly, the fix is cooling, not a different test. Check the cooler mounting, paste, case airflow and fan curve. The safe temperature guide goes into detail.

How to read the results

Prime95 doesn’t display a “PASS” banner. A pass means it ran for your chosen time with no errors and no worker stopping. You can confirm that three ways:

• Look at the workers. Each worker reports its progress in the main window. A worker that has stopped is a failure.
• Open results.txt. It’s in the Prime95 folder and records the test messages. Completed tests typically report that a self-test passed.
• Check your monitor. Clock speeds that sag over time point to throttling, even when there are no errors.

A failure usually appears as a message such as “FATAL ERROR: Rounding was 0.5, expected less than 0.4” or “Hardware failure detected”, and the worker stops. Prime95 compares each calculation’s rounding error to a threshold, so these messages mean a calculation came out wrong. That can be caused by an unstable overclock, too little voltage, memory settings, heat or a failing component.

What to do if it fails

1. Return to stock. Reset BIOS overclocks, undervolts and memory profiles, then retest. Every community troubleshooting thread starts here.
2. Check temperatures. Rerun with a monitor open. If the CPU hits its limit, fix cooling first.
3. Test Small FFTs and Blend separately to narrow down CPU versus memory.
4. Loosen memory. Disable XMP or EXPO and retest.
5. Change one thing at a time. A small voltage bump or a slightly lower clock, then retest.

The full checklist, with the common causes listed in order, is in Prime95 errors explained.

A sensible testing plan

1. Run the online CPU stress test for a minute or two as a sanity check on cooling.
2. Run Prime95 Small FFTs for 15 minutes with a temperature monitor open.
3. Run Prime95 Blend for 30 to 60 minutes.
4. If you’ve overclocked or undervolted, extend the runs to several hours and confirm with a second tool. See Prime95 vs OCCT vs AIDA64 vs Cinebench for how they differ.
5. Finish with real-world use, since synthetic tests don’t catch everything.

Laptops and small-form-factor PCs

Laptops can run Prime95, but their cooling is a fraction of a desktop’s, so they reach thermal and power limits quickly. Plug in the charger, use a hard flat surface, and expect the CPU to throttle or settle at a lower clock. That’s the system working as designed, not necessarily a fault. Treat sustained clock speed and temperature as the result, not a pass or fail. Community advice for laptop testing is the same: run on AC power and monitor per-core clocks and temperatures with a tool such as HWiNFO.

Frequently asked questions

Related guides