Random restarts, blue screens and game crashes that no normal test can catch - OCCT loads your CPU, GPU, RAM and power supply one part at a time and shows you exactly which part is failing, with live temperatures, voltages and error counts.
Your PC freezes, restarts in the middle of work, or shows a blue screen once in a while - and every normal test says everything is fine. That gap exists because normal use never pushes hardware to its limit, so a weak part keeps hiding. OCCT (from OCBASE, free for personal use) closes that gap: it loads one part at a time - CPU, GPU, RAM, graphics memory, power supply - while watching temperatures, voltages and errors live. This guide matches each symptom to the right test, shows safe durations, and teaches you to read the result like a technician.
Download OCCT only from the official site ocbase.com - it is a single portable program, so there is nothing to install; just right-click and run it as administrator so it can read all sensors. The free personal version limits one test run to 1 hour and shows a short wait before a test starts; paid editions remove those limits and add reports. For fault-finding, the free version is enough.
Running every test blindly wastes hours. Start from what the PC actually does wrong:
Test one part at a time. The combined Power test is deliberately the last step, not the first, because it loads everything together and cannot tell you which part failed.
Open the CPU test in OCCT. For a health check on a stock PC, use these settings:
While it runs, OCCT checks every calculation and reports an error count. A healthy CPU finishes with zero errors and no WHEA warnings. Even a handful of errors at stock speed means something real: weak CPU voltage, a marginal motherboard power stage, heat, or a CPU that is genuinely degrading.
The GPU 3D test renders a heavy scene that loads the graphics core. Artefacts (wrong colours, flashing blocks), a driver crash, or a black screen under this test point at the GPU core, its cooling, or its power delivery. Run it for 20 to 30 minutes and watch GPU and hotspot temperatures.
The VRAM test is different: it writes patterns into the graphics card memory and reads them back, cell by cell. A card can pass every game benchmark and still fail here, because benchmarks never fill all the memory. If VRAM errors appear, the card's memory chips or their solder joints are failing - that is a hardware verdict, not a driver problem.
OCCT's Memory test stresses the RAM and the CPU's memory controller together and is much faster at exposing marginal RAM than simply using the PC. Run at least 30 minutes; for a PC that crashes once a day, run the full hour and then confirm with an overnight bootable MemTest86 pass if needed.
The rule is absolute: one error is already a failure. RAM either returns exactly the data that was written, every single time, or it is not trustworthy - there is no acceptable error rate. If errors appear, test one stick at a time to find the culprit.
The Power test runs the CPU and GPU loads together so the power supply faces its highest realistic demand, while OCCT graphs the +12V, +5V and +3.3V rails. Watch for the +12V line sagging far below 12V or jumping around, and for a sudden restart - the classic signature of a PSU that can no longer hold up under load.
Be honest about what it proves: a restart or rail collapse during the Power test is strong evidence against the PSU (or the GPU's power draw), but a pass is not a certificate - a dying PSU can survive one good hour and still fail next week. And never leave this test running unattended.
Write down the test name, duration, error count and peak temperatures for every run. That small record is what turns guessing into diagnosis - and it is exactly the evidence a service centre asks for.
Every test in this guide is shown live in the video - download, setup, CPU, GPU, RAM and PSU testing with real readings.
▶️ Watch Full VideoYes. The personal edition from ocbase.com is free and includes the CPU, GPU 3D, VRAM, Memory and Power tests with live monitoring. Its limits are a 1-hour maximum per test run and a short wait before a test starts. Paid editions mainly remove the time limit and add reports, certificates and commercial use.
On a healthy PC with working cooling, no - the test only creates a heavy load, and modern parts slow down or shut off before heat can harm them. The risk appears on machines that already have a fault, like a failing PSU or a dead fan, which is why you stay with the machine, watch temperatures, and stop immediately if you smell burning or temperatures run away.
For fault-finding, 20 to 30 minutes per test catches most bad parts, and the full free hour is better for faults that appear rarely. The power test should be shorter and always supervised - 10 to 15 minutes tells you what you need.