Problems and testing
Fixing RAM instability on AM5: a checklist
Most memory instability on AM5 comes from the EXPO or XMP profile asking more of the CPU's memory controller than it can give, not from a broken stick. Work through it in order: update the BIOS, check the sticks and slots, test at default speed, then bring EXPO back and ease off one setting at a time. Keep the SoC voltage at or under 1.30V the whole way.
This checklist is for Ryzen 7000 and 9000 desktop CPUs on AM5 boards. Most of it applies to Intel too, but the voltage and clock names differ.
Is it actually the RAM?
Memory problems look like random blue screens (often MEMORY_MANAGEMENT or IRQL_NOT_LESS_OR_EQUAL), games crashing at random points, corrupted downloads, installers failing, or a PC that won't boot after you turn on EXPO.
The quickest check is to turn EXPO or XMP off and use the PC for a day or two. If the crashes stop, your RAM is almost certainly fine and the profile isn't stable on your system. If they carry on at default speed, the cause could be a stick, but it could also be something else entirely: Evetech lists storage corruption, graphics drivers and power faults as causes of crashes that look like memory.
Before you change anything
- Turn off CPU tuning while you test. That means Precision Boost Overdrive (PBO), Curve Optimizer and any undervolt. Evetech's advice is to change one thing at a time and never mix CPU and memory changes in the same step. In one ASUS ROG forum thread, crashes first blamed on the fabric clock turned out to be the CPU's automatic overclocking.
- Back up anything important. Unstable memory can damage files as they're written.
- Write down your current settings, or take photos of the BIOS memory pages.
- Know how to clear CMOS on your board, in case a change stops it booting.
Step 1: Update the BIOS
On AM5 this fixes more memory problems than anything else. In 2023, AMD's AGESA update (the firmware AMD supplies to board makers) added "a cap on SOC voltage at 1.3 V" after some CPUs were damaged, and AMD told users to update. Board makers have kept improving memory training and compatibility since. MSI, for example, warned that Memory Context Restore wasn't stable on older BIOS versions and told owners to update first.
After a BIOS update your settings go back to default, so you'll need to turn EXPO on again. Expect a long first boot while the board retrains the memory; see why DDR5 takes so long to boot.
Step 2: Check the sticks and the slots
- Two sticks go in A2 and B2, usually the second and fourth slots from the CPU. Your manual confirms which.
- Press each stick down until both latches click. A stick that's slightly out causes odd errors as well as missing memory.
- Four sticks are a much harder job for the controller. AMD's official speeds drop sharply with four:
| CPU (example) | 2 sticks, single or dual rank | 4 sticks, single or dual rank |
|---|---|---|
| Ryzen 7 7800X3D (Ryzen 7000) | DDR5-5200 | DDR5-3600 |
| Ryzen 5 9600 (Ryzen 9000) | DDR5-5600 | DDR5-3600 |
Anything above those numbers, including a DDR5-6000 EXPO profile, is technically overclocking. That's normal and usually works with two sticks. With four, it often won't; see 2 sticks vs 4 sticks of DDR5. If you've combined two kits, read can you mix RAM? as well.
Step 3: Test at default settings
Clear CMOS or turn EXPO off, then run MemTest86 from a USB stick for its default four passes. How to run MemTest86, TestMem5 and OCCT covers the setup and how to read the results.
- No errors at default: the hardware is probably fine. Move on to step 4.
- Errors at default: test each stick on its own in the slot your manual recommends for one stick. A stick that fails alone while the other passes in the same slot is faulty.
One catch: a PassMark staff member pointed out that RAM can pass as a single stick and still fail in dual channel. So two sticks that each pass alone but fail together at default speed don't prove either stick is bad. Try them in another AM5 PC if you can.
Step 4: Bring EXPO back, then ease off one step at a time
Turn EXPO on and run 20 to 30 minutes of TestMem5 or OCCT after each change below. Stop at the first step that passes.
- Try the other profile. On ASUS boards, switch between EXPO I and EXPO II. If the kit also has an XMP profile, try that. See enabling XMP and EXPO on each board brand.
- Undo any manual tuning. If you set timings, FCLK or voltages by hand, go back to the plain profile first.
- Check the controller is running 1:1. In ZenTimings, UCLK should match MCLK at DDR5-6000. Leave FCLK on Auto or around 2000 MHz. An unstable FCLK tends to cause odd crashes rather than clean memory errors. Gear 2 and UCLK explained shows where to look.
- Drop one speed step, for example from 6000 to 5800 or 5600, with the profile's timings. Evetech's estimate is that this costs only a few fps in most games.
- Add a little memory voltage (VDD and VDDQ). Go up in small steps of about 0.04V, as a PassMark forum member suggested, and stay at or under about 1.40V for daily use.
- Leave the SoC voltage alone, or raise it only slightly if it's well below the cap. Never go above 1.30V. If a speed only works with more SoC voltage than that, run a slower speed instead.
Changing more than one of these at once makes it impossible to tell what fixed it.
Step 5: Turn Memory Context Restore off while you test
Memory Context Restore (MCR) saves the results of memory training so the board doesn't retrain on every boot. It's useful, but while you're troubleshooting it can keep reusing training results from settings that were never quite stable. In the PassMark forum thread above, the owner of an early AM5 build saw Windows corrupted twice after turning MCR on.
So turn MCR off until everything passes. Once your settings are stable, turn it back on, leave Power Down Mode enabled with it, and test again. If crashes come back, especially after a cold boot, leave MCR off.
Step 6: Check temperatures
DDR5 sticks report their temperature in HWiNFO. Under 50°C is comfortable, and above about 60°C some tight settings start to get unstable. A big graphics card dumping heat into the case can push the RAM warmer during gaming than it ever gets in a memory test. Evetech suggests running the final test alongside a graphics load so the case warms up the way it does in a game.
If the RAM runs hot, a case fan moving some air over the slots usually fixes it.
Step 7: Look for hidden hardware errors
Open Event Viewer, go to Windows Logs, then System, and look for WHEA-Logger entries. These mean the hardware reported an error. On AM5, they can come from the memory, the fabric clock or the CPU's own tuning, so treat any you find like a failed test. Evetech makes the point that a test showing zero errors while Windows logs hardware faults isn't a clean result.
Symptoms and where to start
| What you see | Likely cause | Try first |
|---|---|---|
| No boot or DRAM light stuck on after enabling EXPO | Training can't settle at the profile | Wait five minutes, clear CMOS, update BIOS, then try the other profile or one step lower |
| Crashes only with EXPO on | Profile is too much for the controller | Step 4 |
| MemTest86 errors at default speed | Faulty stick or slot | Test one stick at a time |
| Crashes after an hour or two of gaming | Heat or marginal settings | Step 6, then a longer test |
| Crashes after a cold boot with MCR on | Saved training no longer fits | Step 5 |
| Four sticks won't hold the speed | Four-stick limit | Run slower, or switch to two bigger sticks |
When is it time to RMA?
When one stick fails MemTest86 at default settings, on its own, in the recommended slot, and the other stick passes in that same slot. That's a faulty stick, and most brands will want the whole kit back. See how to RMA RAM. If everything passes at default but EXPO never holds, even one step lower, the kit isn't broken; your CPU's memory controller or board has reached its limit at that speed. Run it at the speed that passes.
The short version
- Update the BIOS first. It fixes more AM5 memory problems than anything else.
- Turn off PBO and undervolts, and use two sticks in A2 and B2.
- If it's stable with EXPO off, the RAM is fine; try the other profile, then one speed step lower.
- Keep memory voltage at or under about 1.40V and never push SoC past 1.30V.
- Turn Memory Context Restore off while testing, and back on once everything passes.
Sources
- AMD, Ryzen 5 9600 specifications (memory speeds)
- AMD, Ryzen 7 7800X3D specifications (memory speeds)
- TechPowerUp, AMD's second statement on Ryzen 7000 issues (SoC voltage cap)
- Unikos Hardware, MSI on Memory Context Restore for AM5
- Evetech, AM5 motherboard instability after enabling EXPO
- Evetech, Faster ways to catch RAM instability with TestMem5 and OCCT
- PassMark forums, AM5 RAM stability and inconsistent MemTest86 results
- ASUS ROG forum, FCLK on Auto crashes with a Ryzen 7800X3D