Memtest86 errors: tyring to troubleshoot my system

[hardway]

Hi,

I have been having some system freezes in Windows 10 and decided to run memtest86 on my new system. It doesn't seem to have any trouble running games or cinebench, or prime95, but it will just freeze up when I have a couple of browser tabs open or open up excel or once when I finished a cinebench run and was opening Edge browser. Never BSOD, just Windows freeze and I have to restart. It always boots up fine too.

System:
CPU: Ryzen 9 9950X
MOBO: MSI MPG X879E CARBON WIFI WITH LATEST BIOS
RAM: G.SKILL TRIDENT Z5 NEO RGB 2X32 GB DDR5 EXPO F5-6000J3036G32X2-TZ5NRW

I have had the EXPO profile (6000 MT/s) enabled and first I ran memtest with it enabled and both sticks of ram (in slots A2 and B2 as described in mobo manual). I got errors so I disabled EXPO and ran memtest again with the memory at 4800 MT/s. Still getting errors. The errors are always the same, test 7, single bit, expected: FF7FFFF Actual: FFFFFFFF, CPU's 4 or 5. The address changes though.

Next I pull out both RAM modules and clean the contacts with rubbing alcohol and a Q-Tip and blow out the slots on the mobo. Then mark the modules with tape, 1 & 2. Then I insert DIMM1 into slot A2 and run memtest again. It gets through 3 passes but fails on the 4th, same error as before.

Then I switch DIMM1 into slot B2 and run memtest again. It gets through 4 passes and passes all the tests, but I decide to run it again because I read I should do 8 passes to have high confidence in the results. This time it makes it through 3 passes but throws 2 errors on the 4th pass (so pass 8 if we're considering the twor runs back to back). Same error both times, but different addresses, one on CPU 4 and one on CPU 5.

Next I insert DIMM2 into slot A2 and run memtest. It throws 1 error on pass 2 and 1 error on pass 4. Same error as always, both CPU 5 but different addresses.

I am now running memtest again with DIMM2 in slot B2, I am assuming it will fail at some point in 8 passes. I attached a table I made with the errors and addresses.

I am not sure what to do. It's always test 7, always the same error, always either CPU 4 or 5, but different addresses. It seems unlikely that I would have 2 bad DIMMS or two bad DIMM slots on the mobo. I don't have any friends with DDR5 ram to borrow from, and this mobo only takes DD5 so I can't use my DDR4 modules to check the slots.

Any suggestions would be very much appreciated!

[David (PassMark)]

It seems unlikely that I would have 2 bad DIMMS or two bad DIMM slots on the mobo.

Correct. And even if both DIMMs were bad, to have them both have bit flips on the same bit would be strange. I mean it could be a systematic manufacturing flaw, so all the DIMMs in a batch from GSkill are bad, but it seems unlikely.

So you need to start to consider the CPU might be at fault.

I would be nice to try some non-GSkill ram however. As there is a chance the SPD data in the sticks was programmed to be too aggressive (e.g. voltage too low, timings too tight). Which I guess also classifies as a systematic flaw, but in the settings and not the hardware.

I would also check if there was any BIOS updates, as from time to time they adjust memory timings and setting in BIOS patches to improve stability.

[Roxxors]

You can go into the BIOS and (if your BIOS supports it) set each cores speed. Back down the speed on cores 4 and 5 a notch or so and run it again. My MB had like 500 fails on "CPU 16" when I had "Sync all cores" enabled. I shut that off, went in and set the speed of each core to what I wanted, then backed core 16 down to 4.8G and left the rest at 5.2G. No more errors.