Diagnostic software splits into two jobs people constantly confuse: monitoring (watching temps, clocks, voltages while the system runs) and stress testing (deliberately pushing a component until it fails or proves stable). You need both, and no single free tool does everything well. Here’s what actually earns a spot on a troubleshooting USB drive.
Monitoring: HWiNFO and HWMonitor
HWiNFO is the one I leave running in the background on every build. It reads sensors most other tools miss: VRM temps on some motherboards, individual CPU core voltages under load, PCIe link speed (useful for catching a GPU that’s silently running at x8 instead of x16 due to a seated-wrong card or a weird BIOS setting), and SSD wear leveling data. The interface is dense and ugly, but the logging-to-CSV feature is genuinely useful if you’re trying to correlate a crash timestamp with a voltage spike.
HWMonitor from CPUID is simpler and friendlier, good for a quick glance, but it misses sensors HWiNFO catches and its min/max tracking resets less reliably across sessions. If you only install one, make it HWiNFO. The learning curve is a weekend, not a month.
Stress testing: Prime95, OCCT, FurMark
Prime95’s Small FFTs mode is still the standard CPU torture test. It pushes thermals harder than almost any real workload, which is exactly the point: if your cooler can’t hold a CPU under 90-95°C on Small FFTs for 15-20 minutes, it will struggle with sustained real-world loads like video encoding or compiling. Note that newer Intel chips (13th/14th gen especially) have had stability issues that look like Prime95 failures but are actually voltage/microcode problems, not cooling problems, so don’t assume a failed run always means bad paste or a bad cooler.
OCCT is the better all-rounder now. It tests CPU, GPU, memory, and power supply stability in one app, and its “Power” test specifically hunts for the kind of transient voltage spikes that trip a marginal PSU’s protection circuits. If a system randomly reboots under gaming load but seems fine otherwise, OCCT’s power test will often reproduce it faster than hours of gaming ever would.
FurMark gets a bad reputation for being “too extreme” and some board partners have throttled or blacklisted it, but as a 5-10 minute gut check for GPU thermal and power delivery problems, it’s still fine. Don’t run it for an hour; you’re not simulating a game, you’re deliberately finding the worst case.
GPU-specific tools
MSI Afterburner (works regardless of GPU brand) is the standard for monitoring clocks, memory temp, and power draw in real time, plus it’s how most people actually apply a manual fan curve. If your GPU is coil-whining, throttling, or crashing only under specific loads, Afterburner’s on-screen display during actual gameplay tells you more than a synthetic benchmark will, because it shows you what’s happening at the exact moment of the problem.
Memory and storage
MemTest86 (the free standalone bootable version, not the Windows-only MemTest) is non-negotiable for a new build or after a RAM upgrade. Run at least 2 full passes, ideally overnight. A single error anywhere means the RAM, the motherboard’s memory controller, or your XMP/EXPO settings are unstable. This is also the tool that catches a lot of “random BSOD after months of fine use” cases that people wrongly blame on Windows.
For storage, CrystalDiskInfo reads SMART data and flags drives with reallocated sectors or rising uncorrectable error counts before they fail outright. CrystalDiskMark tells you if an SSD is performing at its rated spec, useful for catching a drive running in a slower PCIe mode than it should, or a SATA SSD accidentally plugged into a port sharing bandwidth with something else.
| Tool | Best for | Cost | Limitation |
|---|---|---|---|
| HWiNFO | Deep sensor monitoring | Free | Dense UI, steep first use |
| Prime95 | CPU thermal/stability stress | Free | Unrealistically harsh load |
| OCCT | CPU/GPU/PSU combined stress | Free, paid tier available | Paid tier gates some tests |
| MemTest86 | RAM error detection | Free (paid for UEFI extras) | Needs reboot to bootable USB |
| CrystalDiskInfo | Drive health/SMART | Free | Doesn’t catch all failure modes |
When software can’t tell you enough
Diagnostic software only reads what sensors report, and cheap motherboards sometimes report inaccurate VRM or chipset temps. If you suspect a PSU problem specifically, software stress tests help, but a basic PSU tester gives you a hardware-level answer in 30 seconds without touching the OS. Similarly, if you’re chasing a thermal issue and don’t trust onboard sensors, a cheap infrared thermometer lets you spot-check a VRM or heatsink directly. For intermittent crashes that only software monitoring can catch over time, a basic USB flash drive dedicated to bootable diagnostics (MemTest86, a Linux live image for hardware testing outside Windows) is worth keeping in a drawer rather than making one at 1am when something’s already broken.
What to actually run, in order
For a new build: MemTest86 overnight, then Prime95 Small FFTs for 20 minutes, then OCCT’s combined test for another 20, while watching HWiNFO. For an existing system acting up: CrystalDiskInfo first (cheapest check, rules out a dying drive), then HWiNFO logging during the actual activity that triggers the problem, rather than a synthetic test that might not reproduce it at all. Most “random crash” tickets I’ve seen solved weren’t solved by a stress test, they were solved by logging real usage and reading the log afterward.
