Strange memory issue

Associate
Joined
16 Jan 2003
Posts
1,230
Location
by my pc. :D
Hi All,

I've an asrock x570 taichi, with 3700x and 4 x 8GB 8 Pack Dark Pro 3600 ram.

Was having a few issue with a game crashing to desktop so decided to do some diagnostics. Ran Memtest86 (latest version) with the 4 sticks in place. It runs a full test 4 times. I had a single error on three of the full tests.

2020-12-04 14:11:13 - [Data Error] Test: 9, CPU: 14, Address: 77542362C, Expected: FD3081DA, Actual: FF3081DA
2020-12-04 12:08:56 - [Data Error] Test: 8, CPU: 14, Address: 774E5762C, Expected: 13DC2129, Actual: 11DC2129
2020-12-04 10:15:49 - [Data Error] Test: 7, CPU: 12, Address: 76F28B62C, Expected: FDFFFFFF, Actual: FFFFFFFF

Not the same location, during different test sections although does appear to be same bit effected in data? And one pass completed without issue?

Tested each pair on their own and they completed a full test (4 passes) with no issues.

Another oddity is that for some reason one of the dimms does not show any SPD data in CPUZ - yet get's read ok by MemTest and in the bios. Just using this single dimm I'm able to enable XMP in bios and machine boots and loads up windows fine, but no SPD data in CPUZ. The other 3 dimms show SPD info with no issues when installed.

Anybody else seen the SPD issue? Difficult to say whether the memory is faulty as they test out fine in pairs.
 
Further testing - if I put either pair of dimms in slots A1 and B1, then try enabling XMP I get 3 beep codes on reboot - then it goes through the training cycle and finally boots with them at stock speeds (2400).

If I use A2 and B2 and enable XMP machine boots up fine and keeps XMP speeds (which is the recommended slots to use first according to the manual). If I use all 4 slots it's the same - XMP speeds retained.

I guess this is sounding like one of the memory controllers on the CPU is weak, or something effecting channel 1 stability at 3600?

Still can't explain the SPD reading issue though - even tried an earlier version of CPUZ and still the same.
 
Everything at stock, and each pair passes full test in A2 and B2 - don't think it would make any difference testing individually. As stated I can't run any sticks on their own at 3600 in Channel 1 slots only for some reason. At the moment all 4 slots at 3600 is producing the errors in memtest - so I think channel one is unstable at 3600. :(

Last thing to try is all 4 slots at 2400 and see if it still produces any errors. I'll try that tonight.

Still doesn't explain the 1 dimm with the SPD read issue though...
 
I've tried a single dimm in A1 - works ok until I enable XMP in which case I get a 3 beep error. Does it regardless which dimm I use.

And the fact I only get 1 error each pass using 4 dimms doesn't seem to point to a socket issue - it would surely result in loads?
 
as mentioned above try all 4 sticks at 2400 and move up from there. missing spd info not good, if that 1 stick with missing spd info is on its own in the mobo in different slots does it still miss spd info?

The odd thing is that in bios when I enable XMP with only the "suspect" dimm in, it loads the profile fine. It just doesn't seem to show anything in CPUZ, while the other three do. Strange.

If I use just the suspect dimm on it's own the XMP profile works fine in bios when in slot A2, and machine boots - but no SPD info showing for any slots in CPUZ.
 
Last edited:
what about B1? that works ok?

if the have only 1 error, in the same socket no matter what ram you use... I would point at the motherboard.

Trying to recall if I did try just B1 - I know I tried A1 and B1 as a pair and got beep codes when trying to enable XMP, and then tried just A1 with same result. But no beep codes with all 4 in and XMP enabled.
 
https://www.amd.com/en/products/cpu/amd-ryzen-7-3700x

Official ram speed on that spec sheet says upto 3200mhz. If its anything like the chart that was released with first gen ryzen then there is limitations with number of sticks and ranks.

True, the official max speed is 3200, and even has (OC) by anything higher in the manual. Had the system for over a year and only seeing memtest errors now. I guess the mem controller must have weakened a little or something else on the board.

I'll try a run at 3200 with all 4 in, and see how that goes. Not going to do multiple tests and different speeds just to find the fastest speed it will do with 4 sticks stable. If that fails I'll just run on two sticks at 3600. Try 4 sticks again on a 5xxx chip when they are available. :)
 
depends on what you do 4 sticks at slower speed may be considerably better than 2 sticks at faster speed. i use 4 sticks for 64gb and with heavy vm loading its very beneficial to me far more than even half the ram would be even if it ran at 5ghz.

Seems pretty obvious to me - the age old capacity vs speed vs cost decisions. Going to try for speed myself since it helps (a little) for gaming.

Ran a test at 3200 overnight and that passed fine.
 
Don't mind leaving things run overnight - it's not as if I'm using it then.

Think I may have possibly solved the issue. I've upped the memory voltage a little and test seemed to pass at 3600. XMP profile sets mem at 1.35v. Manually set it to 1.375 and seems better. Think max rated voltage for these is 1.4v so have some headroom. Will do a long test tonight.
 
I may check it out - although thought memtest86 was pretty much the defacto standard for memory testing? Using latest version which does seem to do a few more tests than I remember (like a bit fade test).

I ran it overnight with memory at 1.375v and no errors detected. I played the game I had issues with for a few hours last night with no crashes, so looking promising. If this is the cure then guess either memory just needed a bit more power or there is a slight voltage drop on the board?
 
Back
Top Bottom