Give an intermittent problem a repeatable workload
OCCT is a hardware testing and monitoring utility from OCBASE. It can place selected parts of a Windows PC under a defined workload and report detected calculation errors alongside available sensor readings. This helps when a machine crashes only during demanding work or behaves differently after a memory or processor setting has changed. The program creates evidence for investigation; it does not automatically repair unstable hardware.
The verified stable release is 17.1.9, dated October 16, 2026. Personal provides the core tools free for personal, noncommercial use. Its tests are limited to one hour and include a waiting period between runs. Supporter options extend duration and add reports and certificates, but remain personal licences. Testing computers in a commercial or business context requires an appropriate professional edition.
Choose a test that answers your question
CPU + RAM stresses a path involving both processor and memory, while the separate CPU workload concentrates more directly on the processor. Memory testing exercises the RAM it can allocate, not memory reserved by the running operating system. These distinctions matter when interpreting results: an error during a combined test does not identify a particular memory module by itself.
For graphics investigation, 3D Adaptive varies GPU work, while VRAM testing addresses graphics-memory behavior. The developer notes that hidden transfer retries can prevent some transfer-related problems from appearing as reported VRAM errors. A clean result therefore needs to be read within the test's scope. GPU-Z is useful for identifying the adapter before choosing which GPU to exercise on a computer with multiple graphics devices.
Begin with the current configuration recorded
Save open work and note the settings you are trying to evaluate. Start by observing idle readings, choose a relevant workload and watch the first part of the run. Keep cooling unobstructed and use the component manufacturer's limits rather than assuming a single temperature threshold fits all processors and graphics cards. Stop if the computer shows abnormal behavior or cooling cannot cope.
OCCT also has a Power workload combining heavy CPU and GPU demand. It creates a different electrical and thermal situation from a narrow component test. A shutdown under that load is an observation to investigate, not proof that the power supply alone is faulty. Simultaneously changing memory settings, graphics clocks and cooling would make any later comparison difficult to interpret.
Read outcomes as evidence, not guarantees
Record the test name, relevant options, duration and the configuration used. Compare one deliberate change at a time under similar conditions. HWiNFO can provide additional sensor context when a particular reading requires closer examination, subject to the hardware's available sensors. A maximum temperature without the associated workload tells less than the same reading placed on a timeline.
A completed run without errors means no error was reported in that run; it does not certify every possible application or future session. Likewise, a benchmark score measures performance under its own workload rather than proving stability. OCCT is most useful when its results narrow the next investigation and can be reproduced, with licence choice and test scope understood before starting.