Two green memory modules and a USB drive on a blue surface.
Gear / The story

Memtest86+ 9 adds Arm support and new ways to expose memory faults

The free diagnostic gains ARM64 support on compatible firmware, memory-bus stress testing and USB reports. A clean run still has specific limits.

GeekRadarStory date Published 4 min read

Memtest86+ 9.00 brings the free memory-testing tool to supported ARM64 systems and adds tests aimed at failures that can hide behind a clean run of more conventional checks. Released on October 15, it also gains a much simpler way to keep evidence of a diagnostic session: save a report to a USB drive.

The headline needs a qualification. ARM64 support does not mean the same download will test every Arm-powered device. Firmware and platform support still determine whether the tool can boot and reach the hardware it needs to examine.

Arm support has a firmware boundary

The version 9 release information specifies ARM64 systems using UEFI and ACPI. Examples include Snapdragon X computers, supported Raspberry Pi 4 and 5 configurations using community UEFI firmware, and virtual machines under QEMU. That is a defined class of systems, not a promise of compatibility with Android phones or Apple Silicon Macs.

The distinction matters because an instruction set is only one part of a computer. A standalone diagnostic also needs the firmware to expose usable information and services before an ordinary operating system is running. Two machines can execute Arm code while providing very different boot environments.

It is the same general reason that a new Windows-on-Arm laptop platform should be evaluated through actual application and peripheral support, rather than the processor label alone. For Memtest86+, the useful first check is the tool’s documented boot requirements.

Testing the path to memory

The 9.00 release notes list new SIMD-based testing, a memory-bus stress test and an optional Row Hammer test. These additions examine different failure mechanisms; they are not simply more passes through an identical pattern.

The new write tests can bypass the processor cache. A cache normally helps performance by retaining recently used data close to the CPU. In a diagnostic, however, the point is to exercise the memory subsystem rather than repeatedly obtain an answer from that nearer store. Sending work beyond the cache helps put the intended hardware under pressure.

The bus stress test varies activity with idle intervals. As the technical readme explains, changing the load can expose problems in the interface and memory-controller path. This is particularly relevant to DDR5, where correction inside a memory chip does not establish that the entire path between processor and memory is sound.

Row Hammer is different again: it concerns disturbance caused by repeated access to memory rows. Version 9 includes it as test 11, disabled by default. A normal default run therefore should not be described as having performed every available check. Someone comparing results needs to know which tests actually ran.

Keep the result, then interpret it carefully

Pressing F6 can write a report to a FAT32-formatted USB drive. Numbered output files avoid overwriting an earlier result. That makes it easier to preserve a baseline and compare it with a later run after a configuration change, without relying on a photograph of the screen or a remembered error count.

A report is most useful when accompanied by the conditions: the machine, memory configuration, selected tests and the settings under which the run occurred. Change several variables between runs and a better result may be encouraging without identifying which change helped.

The readme also draws an essential diagnostic boundary. Passing a memory test does not certify every component in a computer, while a failure involving the CPU can interfere with the diagnostic itself. Intermittent problems may require longer testing to appear. A red error line provides evidence of a problem under those conditions; it does not automatically identify the single part that should be replaced.

Version 9’s practical improvement is broader coverage with better records. Its value is highest when the result becomes part of a controlled investigation: establish what was tested, retain the report, and change one relevant condition at a time. The download and boot instructions remain available through the project’s official site.