Question for y’all with a lot more experience with hardware. Built my first DDR5 system over the holiday and decided to memtest86+ it, got strange results and looking for any input on tests I should try or where to go from here (thinking I should return the whole system at this point.):
Running stock XMP settings (6000) on dual channel RAM
LGA 1700 system meant for gaming and AI tinkering use case.
Ran memtest86+ for 3 days - no errors reported
Skipped a day of testing, then tested with four passes another day, got a single error in one pass, flipped bit. (Socket B2)
Was able to replicate this again. Running many tests, gives no errors. Then, waiting a day, gives a single bit flip error, in B2 again.
Tried switching RAM slots (flip A and B), same issue where it’s good for a while, then gives one error.
Does not crash when running cinebench r23 for 12 hours straight. No errors in syslog either (windows 10)
Has anyone run into this before where the system doesn’t crash, passes a lot of memtest86+ passes, but after resting for a day, fails a single test?
That error is a little suspicious, if you don’t live at a high altitude I’d try running the memtest again at a reduced RAM speed and see if that remedies the memtest results. a memory error that seldom doesn’t usually translate into Windows crashes very often, but if/when it does it’ll be a pain to deal with.
Welcome to that .1% of instability! Can probably be fixed with a slight voltage increase. I’d do what twin_savage said and test at 4800 just to see if it still shows up (if it does, return or RMA the RAM). If it doesn’t then you can get 6000 stable with a little bit of tuning. But I absolutely do not recommend using the system with errors detected. If errors show up on memtest then eventually they’re going to affect the OS, drivers, and data integrity on your hard drive.
Small update for y’all. Lowered the RAM speed to 4800 like you both suggested. Ram memtest86+ the past few days. No reported errors, but the program did act a bit odd:
During one run the RAM temperature that normally shows as 0C or N/A became a negative number (bit flip perhaps?)
During longer overnight runs, the ‘press any key’ to continue screen would take me to a black screen instead of further info when a key was pressed.
Think these could be both memory or motherboard issues presenting themself through Memtest86+.
I did also run Prime95 in RAM/Memory Controller test mode for 1 hour and no errors were reported.
Update2: Ran Memtest86+ 12 more passes, no errors at 4800. Also ran Prime95 for 6hrs in RAM / Memory controller test mode, no crashing or errors reported. Perhaps it’s that XMP (6000) is not stable at stock voltage on this RAM?
What model CPU, motherboard, and Ram are you using? And are you overclocking anything on the system?
Socket 1700 can be picky with certain memory controller, motherboard, and memory combinations wanting particular voltage settings that can sweet spot; meaning that more voltage is sometimes worse or even causes regression.
CPU: intel core i5 13500
Motherboard: MSI z690 force Wifi
RAM: Corsair Vengeance RGB Pro 64 GB (2 x 32 GB) DDR5-6000 CL30
*disabled fast boot, and I physically turn off / unplug before each Memtest86 run since I read hard/cold boots are more stressful for RAM, and I want to make sure they're stable.
Not overclocking anything else on the system (well, atleast not specifically, who knows what motherboard vendors like to do under the hood.)
I did try XMP 6000, increasing both DDR5 voltages by .005, but was less stable in Memtest86 than at stock (3 flipped bits in slot B2)
Right now I’ve got a second set of sticks that I’m trying at XMP stock
voltages, just to rule out if it’s a bad stick.
@GTwannabe that’s next on my list if this new set fails too (especially if the errors come from B2 again.)
The i5 3500 is just a rebadged Alder Lake part, which is why it’s only officially rated for DDR5 4800. That you are trying to run 64GB is also part of the problem.
You are not going to be capable of running 6000 on that processor, or at least not without a great deal of time spent manually voltage & timing tuning. I’d suggest sticking to 4800.
Yeah, I can see on ARC that it’s rated for 4800, perhaps I’ll leave it stock then; my second kit is stable at stock speeds but gives a single error when at XMP… should still be faster than the 1600 ddr4 on my old desktop
Just curious what’s the issue with running 2 sticks of 32GB? The arc page seems to indicate its max ram capacity is 192 GB. Is there some trickery going on with larger sticks using more than just 2 channels?
More capacity means more banks of memory, which increases the electrical load on the memory controllers. Just as clocking the memory higher stresses the IMCs so does memory capacity, and unfortunately you’re wanting to do both on a first-gen DDR5 memory controller. Using 4 modules instead of 2 also adds to the stress on the IMCs.
The max memory capacity is a rather bogus spec, even running 192GB on an AMD processor would force you to run at a very low speed if it worked at all. If I remember the specifications Wendell was having to run slower than 4000 just to avoid errors with 128GB of DDR5.
Friend of mine just put together their first major desktop in 15 years, but surprised me going for 64GB… but he couldn’t even get that stable on his 7800X3D, windows was throwing errors during install until I helped him, but he’s not even able to get 64GB stable at 5600.