How to Test DDR5 XMP and EXPO Stability with TestMem5 on Windows

Where to get TestMem5, and why provenance matters​

The shattered.io walkthrough points readers to a GitHub organization called tm5-memory-test as the official source. Our check did not support that. The organization page reads like a README for the tool. It lists one repository, a .github profile repo, and no release downloads that we could see.

The repository with a real release history is CoolCmd/TestMem5. Its release list shows version 0.13.1 as the latest, with a note about moving part of the resources from TM5.EXE into TM5.DLL. That note was written in Russian and was aimed at antivirus false positives. The earlier 0.13 release is described as the first version of the program.

Practical consequences:

• Download TM5 from the CoolCmd/TestMem5 repository's Releases page. Do not use lookalike mirrors.
• An Overclock.net forum user reported that Defender flagged the archive as a generic trojan and TM5.exe as a PUP. That report is old and anecdotal. It does show that stress tools often trigger antivirus alerts.
• If you get an alert, confirm you downloaded from the right repository before doing anything else. Don't add an exclusion as a reflex.
• Released archives have included a few small .cfg files. One urlquery scan of the 0.13.1 archive listed Extreme @ anta777, Heavy @ anta777 and Universal 2 @ LMhz configs. The earlier archive included 1usmus v3 and Absolut @ anta777.
• Version 0.13.1 is what the releases page shows as latest as of our search. Check the page for anything newer.

What a TM5 pass does and doesn't prove​

XMP and EXPO are memory overclocking profiles. Intel's XMP page says profiles hold preset speed, timing and voltage combinations tested by the manufacturer. The same page says XMP certified modules were tested on specific motherboards and specified processors. It also says that using XMP beyond Intel's specifications may void the processor warranty.

Your CPU's memory controller, your board's layout and BIOS, and your DIMM population all differ from the vendor's test bench. A profile that works there can fail on your machine.

The framing in the project README is a good model for reading results:

• Errors straight away mean the profile is unusable.
• An error after a long run is still a fail for daily use.
• A clean pass of a strict config gives reasonable confidence in that setting.
• A pass of a tiny config is only a smoke test.
• A crash or reboot with no error line also counts as a fail.

A pass covers only the machine, BIOS settings, config and run length you used. It is not a vendor certification, and it doesn't transfer to another PC.

Step-by-step: validating a DDR5 profile​

1. Record what you are testing​

Before you change anything, photograph the BIOS memory page. Capture the profile name, frequency, primary timings, voltages and which slots are populated. Then read the same values in Windows using CPU-Z or HWiNFO. Test only the configuration you intend to keep.

2. Reduce the variables​

• Update the BIOS only if you have a reason to. If you do, use the board maker's stable release.
• Load optimized defaults, then enable profile 1 of your EXPO or XMP set.
• Turn off CPU overclocks and undervolts while testing memory, so a crash isn't misattributed.
• Leave other memory-related settings on Auto where you can.

Memory training on the first boot after a change can take a while, so wait before assuming a hang.

3. Prepare Windows​

• Close browsers, launchers and games. The README asks for a PC that is otherwise idle.
• Keep the case closed as you normally use it, since heat matters.
• On a laptop, plug in. The README notes that power limits change the test.

4. Extract and launch TM5​

Extract the archive to a folder you own, such as C:\Tools\TM5. Open TM5 and load a config. Confirm which config is active before you start the clock. A result is hard to interpret if you don't know what actually ran. Write down the config's name next to your speed, timings and voltage.

We could not verify that TM5 requires administrator rights or that running without them reduces the memory it tests. If you see odd allocation behavior, treat that as something to investigate rather than a documented rule.

5. Run in stages and stop at the first error​

The README says not to start a three-hour run on a profile that errors in the first minute. A sensible approach:

1. Start with a short run to catch gross failures.
2. If it is clean, run a strict config for a longer period. This is the sign-off run.
3. For a workstation or a profile you're tight on, add a longer run with the case closed.

The shattered.io guide suggests 15 minutes, then 60 minutes or more, then three hours, then overnight. Those figures are that publication's own workflow, and community guides repeat the one-hour figure. Neither the TM5 README nor the vendors set a required duration. Treat them as a reasonable ladder, not a standard.

6. Treat any failure as a failed candidate​

If TM5 reports an error, stop. Write down the elapsed time, the pattern or error shown, and the settings. Then change one thing and repeat the same test. The README's guidance is to drop the speed or tight timings, reboot and test again. It says a kit that errors at its advertised profile needs a more conservative setting or a return.

The README also says to raise voltage only inside the kit's documented limits and not to invent voltages from a forum. Because DIMM voltages are board-specific, this guide gives no numbers. Use your kit label and board manufacturer's limits.

A sensible order for changes, from least to most risky:

1. Remove other tweaks, then re-select profile 1.
2. If you run four DIMMs, try two. Vendors warn that four-DIMM operation can limit achievable speed, though we couldn't confirm the details from a primary source.
3. Step the frequency down one notch.
4. Loosen primary timings slightly.
5. Adjust voltages only within published limits.

7. Cross-check in Windows​

TM5 has no memory-error reporting beyond its own window, so look at the system too. Open Event Viewer, then Windows Logs, then System, and look for WHEA-Logger entries and unexpected-restart events during your test window.

Microsoft documents WHEA as the Windows architecture for hardware error reporting and recovery. A WHEA entry is a flag, not a verdict. It doesn't prove the RAM caused the event, since CPU, cache, PCIe and other hardware errors also go through WHEA. Look at the timestamp and the event details. An entry that coincides with a TM5 run on a memory profile is a good reason to be suspicious and to retest at a more conservative setting.

A PowerShell query can list recent WHEA-Logger events in the System log:

$since = (Get-Date).AddHours(-6)
Get-WinEvent -FilterHashtable @{
  LogName='System'
  ProviderName='Microsoft-Windows-WHEA-Logger'
  StartTime=$since
} -ErrorAction SilentlyContinue |
Select-Object TimeCreated, Id, LevelDisplayName, Message | Format-List

An empty result is the outcome you want. This is a diagnostic aid, not proof of stability.

8. Test real workloads and a second tool​

One tool is one opinion. Add another memory test, such as MemTest86+ booted from USB, and use the PC for what you built it for. The README is clear that if games still crash after a strict pass, the GPU and game files are worth examining, since memory isn't the only cause.

A caution about the audit script in the original guide​

The shattered.io guide includes a Python script that records DIMM details and WHEA and Kernel-Power events, then prints a PASS or FAIL. It is a rough logging aid and no more. It never runs TM5 or reads TM5's result. It could print PASS if it finds DIMMs and no events, even though no stress test ran.

Its look-back window may also include unrelated events. Its speed check compares your target with Windows' reported configured speed, and the unit and reporting can vary by platform. Verify the field on your own hardware before trusting it. If you use something like it, treat it as an inventory log and not a pass certificate. A paper log of config name, settings, duration and outcome does the same job.

Common failure points​

The bottom line​

TM5 is a quick, strict way to find out whether your DDR5 profile misbehaves on your build. Its value lies in treating any error as a fail and recording exactly what you tested. A clean strict run raises confidence for that setting only. Combine it with a clean System log, a second tool and your real workloads before you call a profile stable.

The source guide also cites memory prices and various BIOS and AMD EXPO updates. We could not corroborate those figures independently, so we left them out. Download TM5 from the CoolCmd/TestMem5 repository's Releases page.