A stress test runs your PC's hardware at full capacity for an extended period to see if it crashes, overheats, or produces errors. The idea is simple: if your system survives 30 minutes at 100% load without problems, it will handle normal use without issues.
You'd run a stress test after building a new PC (to verify all components work together), after overclocking (to confirm the new settings are stable), when diagnosing random crashes or freezes (to isolate which component is failing), or after replacing a CPU cooler (to confirm the thermal paste application is good).
The process takes about an hour total, uses free software, and can save you from data loss, corrupted files, or hardware damage down the line.
Before you start: safe temperature limits
Stress testing pushes your hardware to temperatures it wouldn't normally reach. You need to monitor temperatures throughout the test and stop immediately if they exceed safe limits.
According to Intel and AMD's published specifications:
CPU temperatures:
- Under 80°C during a stress test: good
- 80-90°C: acceptable for short periods, check your cooling
- 90-100°C: too hot, stop the test and improve your cooling
- 100°C+: thermal throttling territory, the CPU will slow itself down to avoid damage
GPU temperatures:
- Under 85°C: good
- 85-95°C: acceptable for most GPUs under full load
- 95°C+: too hot, improve case airflow or GPU cooling
RAM: Doesn't typically overheat unless overclocked with high voltage. No temperature monitoring needed for stock RAM.
Monitoring tool: Download HWiNFO (free) before you start any stress test. It displays real-time temperatures for every sensor in your system. Keep it running in a second window while the stress test runs. HWiNFO logs data over time, so you can review the maximum temperature reached after the test finishes.
CPU stress test tools
Prime95 (free, Windows/Mac/Linux)
The most widely used CPU stress test since 1995. Prime95 calculates prime numbers using FFT (Fast Fourier Transform) operations that put extreme load on the CPU and memory subsystem. Originally a distributed computing project for finding Mersenne prime numbers, it became the standard stability test because the math is so demanding that even minor instability causes calculation errors.
How to use it:
- Download Prime95 from mersenne.org
- Launch it and select "Just Stress Testing" when prompted
- Choose "Blend" mode (tests both CPU and RAM) or "Small FFTs" (CPU-only, maximum heat)
- Let it run for 30 minutes
What to look for: If Prime95 reports a "FATAL ERROR" or a worker stops, your system is unstable. If you're overclocking, reduce your overclock or increase voltage. If you're at stock settings, check your cooling.
How long: 30 minutes catches most instabilities. Overclockers typically run it for 1-4 hours for extra confidence. Running it overnight (8+ hours) is unnecessary for most users and puts unnecessary wear on components.
OCCT (free for personal use, Windows)
OCCT (OverClock Checking Tool) tests CPU, GPU, RAM, and power supply in one application. Its CPU test is less aggressive than Prime95's Small FFTs but more representative of real-world workloads. OCCT also has a built-in automatic shutdown that stops the test if temperatures exceed a threshold you set.
How to use it:
- Download OCCT from ocbase.com
- Select the CPU test tab
- Choose "Medium" data set for a balanced test
- Set the duration to 30 minutes
- Click Start
Why choose OCCT over Prime95: OCCT's temperature protection is automatic. It also tests the power supply (through voltage monitoring), which Prime95 doesn't. If your PSU can't deliver stable power under load, OCCT catches it. According to the tool's developer, OCCT's "Medium" test catches 95% of the instabilities that Prime95's 24-hour test finds, in 30 minutes.
Cinebench R24 (free, Windows/Mac)
Cinebench is more of a benchmark than a stress test, but its multi-core test is useful as a quick 10-minute stability check. It renders a 3D scene using all CPU cores. If the render completes without errors and your score matches what reviewers report for your CPU model, your system is stable for normal workloads.
Cinebench won't catch marginal instabilities the way Prime95 or OCCT will. Use it as a first pass, not a definitive stability test.
GPU stress test tools
FurMark (free, Windows)
FurMark renders an extremely complex fur texture in a spinning donut shape that pushes your GPU to 100% utilization and maximum temperature. It's the most aggressive GPU stress test available. If your GPU survives 15 minutes of FurMark without artifacts (visual glitches), crashes, or driver resets, it's stable.
How to use it:
- Download FurMark from geeks3d.com
- Set the resolution to your monitor's native resolution
- Check "GPU temperature alarm" and set it to 95°C
- Click "GPU Stress Test"
- Run for 15 minutes
What to look for: Visual artifacts (colored squares, flickering lines, missing textures) mean the GPU is unstable. A black screen followed by "display driver has recovered" means the GPU crashed. Both indicate a problem with the GPU itself, its overclock settings, or inadequate power delivery.
Warning: FurMark generates more heat than any game or application your GPU will ever encounter. Some GPU manufacturers have stated that FurMark's load pattern is unrealistic. If your GPU hits 95°C in FurMark but stays under 85°C in actual games, the GPU is fine for real use. FurMark is a worst-case scenario test.
Unigine Heaven/Superposition (free versions, Windows)
If FurMark is too aggressive for your comfort, Unigine's benchmarks provide a more realistic GPU stress test. They render detailed 3D environments (a floating city in Heaven, a futuristic apartment in Superposition) that represent actual gaming workloads better than FurMark's synthetic fur donut.
Run the benchmark in a loop for 30 minutes. If the system is stable and temperatures stay within limits, your GPU handles real games.
RAM stress test tools
MemTest86 (free, boots from USB)
MemTest86 is the standard RAM stability test. It runs outside of Windows (boots from a USB drive) so it can test all of your RAM without the operating system using part of it.
How to use it:
- Download MemTest86 from memtest86.com
- Create a bootable USB drive using the included tool
- Restart your PC and boot from the USB drive
- MemTest86 starts automatically and runs through multiple test patterns
- Let it complete at least 1 full pass (takes 30-60 minutes depending on RAM size)
What to look for: Any errors (shown in red) mean your RAM is faulty or your RAM overclock (XMP/EXPO profile) is unstable. Even a single error in a MemTest86 run means the RAM is not stable. One error can cause file corruption, blue screens, or application crashes during normal use.
Windows Memory Diagnostic (built into Windows)
If you don't want to create a bootable USB drive, Windows has a built-in memory test. Search for "Windows Memory Diagnostic" in the Start menu, choose "Restart now and check for problems." The test runs during the next restart. It's less thorough than MemTest86 but catches major issues.
Step-by-step: stress testing a new PC build
If you've just built a PC or installed new components, here's the full testing sequence.
Step 1: Update everything. Install the latest BIOS/UEFI, GPU drivers, and Windows updates. Outdated firmware can cause instability that has nothing to do with hardware.
Step 2: Run MemTest86 first (1 pass, ~45 minutes). Test RAM before anything else. Faulty RAM causes random crashes that look like CPU or GPU problems. If MemTest86 shows errors, fix the RAM issue first (reseat the sticks, test one stick at a time, disable XMP/EXPO).
Step 3: Run a CPU stress test (30 minutes). Use OCCT or Prime95. Monitor temperatures with HWiNFO. If the CPU stays under 85°C and the test completes without errors, the CPU and cooler are good.
Step 4: Run a GPU stress test (15-30 minutes). Use FurMark or Unigine Heaven. Watch for artifacts, driver crashes, and temperatures above 95°C. If clean, the GPU and its power delivery are stable.
Step 5: Run a combined test (15 minutes). Use OCCT's "Power Supply" test, which loads CPU and GPU simultaneously. This tests whether your power supply can deliver enough wattage when everything is running at full load. An undersized or failing PSU will cause crashes here that don't appear in individual component tests.
Total time: about 2 hours. After this sequence, you can be confident your system is stable for daily use, gaming, and workstation tasks.
When to stop a stress test
Stop immediately if:
- CPU temperature exceeds 95°C
- GPU temperature exceeds 100°C
- The system blue-screens (BSOD)
- You smell burning (extremely rare, but shut down and unplug immediately)
- HWiNFO shows CPU core voltage dropping under load (PSU issue)
A crash during a stress test doesn't mean your hardware is damaged. It means the current configuration isn't stable. For overclocked systems, reduce the overclock. For stock systems, check cooling (thermal paste, fan curves, case airflow) and power delivery (PSU wattage, cable connections).
FAQ
Is stress testing safe for my PC?
Yes, when done correctly. Modern CPUs and GPUs have thermal protection that shuts them down before damage occurs. Monitor temperatures throughout the test and stop if they exceed safe limits (95°C for CPU, 100°C for GPU). Don't run stress tests for longer than recommended.
How long should I run a stress test?
For CPUs: 30 minutes catches most instabilities. Overclockers run 1-4 hours. For GPUs: 15-30 minutes. For RAM: at least 1 full pass of MemTest86 (30-60 minutes). Running tests overnight is unnecessary for most users.
What's the best free stress test for PC?
OCCT for an all-in-one solution (CPU, GPU, RAM, PSU). Prime95 for CPU-specific torture testing. FurMark for GPU maximum load. MemTest86 for RAM. All four are free for personal use.
My PC crashed during a stress test. Is it broken?
Not necessarily. A crash means the current configuration isn't stable under full load. For overclocked systems, reduce the overclock. For stock systems, check your cooling (thermal paste, fan mounting, case airflow) and power supply (sufficient wattage, secure cable connections).
Should I stress test a new laptop?
A quick 10-minute Cinebench run is reasonable to confirm the laptop performs as advertised. Full stress testing (Prime95, FurMark) generates excessive heat in a laptop's confined space and can trigger aggressive thermal throttling. Laptops are designed around thermal limits, so sustained stress testing has less value than it does for desktops.
How often should I stress test my PC?
Test after building or buying a new PC, after overclocking, after replacing cooling hardware, and when diagnosing unexplained crashes. For a stable system that hasn't changed, yearly testing is more than enough. Some experts, including Avira's technical team, recommend once a year at most for unchanged hardware.
