Every Alexander PC undergoes a multi-stage validation process covering memory integrity, CPU calculation stability, GPU and VRAM stability, combined system power, thermal performance, storage health, real-world workload performance, connectivity, restart and sleep behavior, and Windows hardware-error logs. That comes to roughly 14 hours of measured testing per system, plus four full passes of MemTest86. A system does not ship if any test reports an error or if its performance falls outside our approved range.
Before we stress anything: pre-test inspection
Stress testing a machine that was not built correctly in the first place only tells you that it fails. So the bench work comes first. On every system we:
- Verify every cable and connector
- Check GPU support and seating
- Confirm pump connection and RPM
- Confirm every fan spins and follows the intended curve
- Verify RAM is installed in the correct motherboard slots
- Confirm CPU cooler mounting pressure and contact
- Inspect for loose screws or foreign material
- Update BIOS only to an internally approved, validated version — usually the newest non-beta release
- Load optimized defaults
- Enable the intended XMP or EXPO profile
- Enable Resizable BAR
- Confirm TPM, Secure Boot and UEFI configuration
- Install current chipset, network, audio and GPU drivers
- Update SSD firmware where appropriate
- Confirm Windows reports no unknown devices
We test the machine you actually receive
We test the PC in the configuration delivered to our clients, with the intended memory profile enabled. Testing at JEDEC defaults and then enabling XMP or EXPO afterward defeats much of the purpose — the configuration that passed validation would not be the configuration in the box. Memory instability in particular tends to show up only once the rated profile is running.
The test schedule
Every build runs the following stages. Times are minimums, and any stage that surfaces something unusual gets extended.
- MemTest86 — 4 full passes. Memory integrity. Errors here masquerade as random crashes for months if they are not caught on the bench.
- Prime95 Small FFTs — 60 minutes. Maximum sustained CPU and cooling load.
- Prime95 Blend — 120 minutes. CPU together with the memory controller, where an unstable memory profile shows itself.
- Cinebench CPU loop and comparison — 60 minutes. Sustained multi-core performance measured against the expected range for that hardware.
- FurMark — 60 minutes. GPU stress and thermal behavior.
- FurMark and Prime95 simultaneously — 30 minutes. CPU and GPU loaded at once, which is where marginal power delivery and airflow reveal themselves.
- OCCT VRAM — 60 minutes. Video memory error detection.
- OCCT adaptive — 60 minutes. Shifting load patterns, which catch faults that steady-state testing can miss.
- Combined system power test — 30 minutes. Full-system draw under simultaneous load.
- SSD and device verification — approximately 30 minutes. Drive health and performance, every advertised port and device.
- Boot, sleep and log review — approximately 30 minutes. Cold boot, restart and sleep behavior, and a review of Windows hardware-error logs.
- Gaming benchmark loop — 300 minutes. Five hours of sustained real-world workload, because plenty of faults only appear after hours of actual use.
Sensors are logged with HWiNFO throughout. Drive health is verified with CrystalDiskInfo or the manufacturer's SSD utility, storage performance with CrystalDiskMark or an equivalent controlled benchmark, and hardware faults are reviewed in Windows Reliability Monitor and Event Viewer.
Temperature standards
We deliberately avoid publishing one universal maximum temperature, because modern CPUs intentionally operate differently by platform. A number that means one thing on one platform is misleading on another. Instead we define standards internally by component family, against these principles:
- No component may exceed its manufacturer-defined safe operating limit
- No unexpected thermal throttling is acceptable
- Temperatures must be consistent with our validated reference builds
- Cooling must retain reasonable margin under realistic combined load
For example: a CPU briefly touching its designed thermal target is not automatically defective — that is how the part is engineered to behave. But a premium liquid-cooled system sitting at that ceiling while losing substantial clock speed is a signal, and it usually points to inadequate cooler mounting, poor airflow, an incorrect pump curve, or excessively aggressive motherboard power settings. Catching that distinction is the entire point of this stage, and it is why we compare each build against expected ranges by platform rather than an oversimplified company-wide rule about specific temperatures.
A build passes only if
- MemTest86 reports zero errors
- Prime95 reports zero worker or calculation errors
- OCCT reports zero detected errors
- There are no WHEA hardware errors
- There are no GPU-driver resets
- There are no unexplained reboots, hangs or black screens
- Temperatures remain within our approved limits
- The system does not exhibit unintended thermal throttling
- Benchmark performance falls within our expected range for that configuration
- Every advertised port and device works
- Sleep, restart and cold boot work consistently
- SSD health and performance are normal
- Fan, pump and RGB settings persist after restart
If any one of those fails, the machine does not ship. It goes back to the bench, the cause is identified and corrected, and the system is validated again before it is cleared.
Why we publish this
“Stress-tested” is a claim every builder makes. This page is what ours actually means, written out so you can hold us to it — and so you can ask anyone else you are considering for theirs.
Every system we build is also backed by our warranty and lifetime service policy, and supported by the same people who built and tested it.
Frequently asked
How long does Alexander PCs test each computer?
Roughly 14 hours of measured testing per system, plus four full passes of MemTest86, which runs longer on larger memory configurations. That includes a 5-hour gaming benchmark loop, 3 hours of Prime95 CPU stress across two profiles, 2 hours of OCCT error detection, an hour and a half of GPU and combined CPU/GPU load, and time set aside for storage verification, boot and sleep testing, and log review.
Does Alexander PCs test every computer before it ships?
Yes. Every Alexander PC goes through a multi-stage validation process covering memory integrity, CPU calculation stability, GPU and VRAM stability, combined system power, thermal performance, storage health, real-world workload performance, connectivity, restart and sleep behavior, and Windows hardware-error logs. A system does not ship if any test reports an error or if its performance falls outside our approved range.
What software does Alexander PCs use to test PCs?
MemTest86 for memory integrity, Prime95 for CPU stress and calculation validation, OCCT for CPU, GPU, VRAM and combined-power error detection, FurMark for GPU stress and combined CPU/GPU load, Cinebench for system performance validation, 3DMark or another game-style benchmark for real-world performance, HWiNFO for sensor logging, CrystalDiskInfo or the manufacturer's SSD utility for drive health, CrystalDiskMark or an equivalent controlled benchmark for storage performance, and Windows Reliability Monitor and Event Viewer for hardware-error logs.
Is the PC tested with XMP or EXPO enabled?
Yes. We test the system in the exact configuration it is delivered in, with the intended XMP or EXPO memory profile enabled. Testing at JEDEC defaults and then switching on XMP or EXPO afterward defeats much of the purpose of testing, because the configuration that was validated is not the configuration the customer receives.
What is checked before stress testing begins?
A full physical and firmware inspection: every cable and connector, GPU support and seating, pump connection and RPM, every fan spinning and following its intended curve, RAM in the correct motherboard slots, CPU cooler mounting pressure and contact, and a check for loose screws or foreign material. We then update the BIOS only to an internally approved, validated version, load optimized defaults, enable the intended XMP or EXPO profile, enable Resizable BAR, confirm TPM, Secure Boot and UEFI configuration, install current chipset, network, audio and GPU drivers, update SSD firmware where appropriate, and confirm Windows reports no unknown devices.
What temperatures does Alexander PCs consider acceptable?
We avoid publishing one universal maximum, because modern CPUs intentionally operate differently by platform, so a single number is misleading. We define standards internally by component family against four principles: no component may exceed its manufacturer-defined safe operating limit, no unexpected thermal throttling is acceptable, temperatures must be consistent with our validated reference builds, and cooling must retain reasonable margin under realistic combined load. A CPU briefly touching its designed thermal target is not automatically defective; a liquid-cooled system sitting at that ceiling while losing substantial clock speed is a signal, and usually points to cooler mounting, airflow, pump curve, or motherboard power settings.
What happens if a system fails a test?
It does not ship. The system goes back to the bench, the cause is identified and corrected, and the machine is validated again before it is cleared for shipping.