Instability with 4xDIMM Vengeance LPX 2666 on Gigabyte Z490 UD

[Ahnion]

I recently went from an i5 6600K on a Gigabyte Z170 HD3P to an i5 10600K on a Gigabyte Z490 UD. I was using four sticks (32 GB total) of Corsair Vengeance LPX 2666 MHz (CMK16GX4M2A2666C16 ver3.20) on the previous system. They ran fine with XMP settings. With the new CPU and motherboard, however, the memory just will not run stable with four sticks even with standard SPD settings.

 

Currently, I'm running three sticks with XMP settings and the system is stable. (I've done a bunch of passes in MemTest86 with no errors.)

 

Having done a fair amount of testing, I've concluded that if there is a way to get all four sticks running stable with this motherboard, it is well beyond my present knowledge level, so I'm turning to this forum in the hope that someone might know how to get there.

 

Here's a list of stuff I've tried and concluded:

 

• All sticks check out individually, regardless of slot. One, two (either pair) or three sticks run merrily with XMP settings and no errors in MemTest86. Problems only arise with all four sticks installed.
   
  
• The system is currently running at default clock for the 10600K and temperatures are good.
   
  
• Motherboard BIOS version is F6b, which is currently the most recent.
   
  
• Regardless of the order in which the sticks are installed, MemTest86 errors always occur in the 5xxxxxxxx memory range and the actual return value is always 40 off from the expected value.
   
  
• Errors are few and come sporadically in MemTest86 passes. They do not occur more frequently under load tests.
   
  
• I've had the DRAM voltage as high as 1.24 V with no difference. (SPD says 1.20 for all profiles.)
  

 

I realise that I'm asking for something that's not technically certified, and if three sticks is what I can get to work, well... 24 GB will do for now. Seems silly not to ask, though.

[LeDoyen]

Well you lose dual channel kinda so it will be quite a hit on performance.

 

It would be good to try running just one stick and try all 4 sockets on their own. It is somewhat common to have solder joint failures on one socket.

If you have a defective socket, you could change your sticks till kingdom come, won't work :)

If there's a position that always causes memory errors then you know you have to ask for a motherboard exchange wherever you bought it.

 

If it works fine on 3 sockets and not on 4, i am guessing it's not a problem with CPU seating or bent pins on the socket. You'd typically get errors on 2 sticks out of 4 if that was the case.

[Ahnion]

Well you lose dual channel kinda so it will be quite a hit on performance.

 

It would be good to try running just one stick and try all 4 sockets on their own. It is somewhat common to have solder joint failures on one socket.

If you have a defective socket, you could change your sticks till kingdom come, won't work :)

If there's a position that always causes memory errors then you know you have to ask for a motherboard exchange wherever you bought it.

 

If it works fine on 3 sockets and not on 4, i am guessing it's not a problem with CPU seating or bent pins on the socket. You'd typically get errors on 2 sticks out of 4 if that was the case.

 

Cheers for the answer.

 

I was probably unclear. I've tested using a single stick in each slot (as well as varying which stick I used.) None of them get errors.

 

Also, I may be missing something, but from what the system reports it is running dual channel with three sticks. CPU-Z says dual channel, as does HWiNFO, and the bandwidth reported in MemTest86 is only a few percentages lower than with four sticks installed.

[LeDoyen]

okkkk yes i misunderstood then.

I don't know the details about dual channel. could be it runs 16gb in dual and 8 in single.

 

Bit lost then ^^' 2666 is the base frequency for the memory controller.. could be a dodgy stick but hard to say without spares.

4 sticks will always be inherently less stable than 2 for overclocking but again.. not a high frequency to start with.

[Corsair Notepad]

The other big issue at hand here is you are mixing two kits of memory together and expecting them to work as a single kit. The memory is only validated and guaranteed to work in the configuration they are packaged and sold as and mixing kits can lead to issues like this. Just because your previous system didn't have issues with it doesn't mean your new system can't be effected by this sort of compatibility issue.

 

Have you tested the individual kits in both the new system and old system after a CMOS clear while installing the modules in only the 2nd and 4th slot from the CPU?

[Ahnion]

okkkk yes i misunderstood then.

I don't know the details about dual channel. could be it runs 16gb in dual and 8 in single.

 

Bit lost then ^^' 2666 is the base frequency for the memory controller.. could be a dodgy stick but hard to say without spares.

4 sticks will always be inherently less stable than 2 for overclocking but again.. not a high frequency to start with.

 

I don't have in-depth knowledge either, but it's scampering along happily, so... you know... gift horses and all that.

 

...and yeah, it's strange. I might be able to borrow another set in the coming week, just to test it out. At least it'd be more information.

Also, the fact that there are just as many errors running in standard SPD mode (2133 MHz with slightly looser timings) sort of indicates that the problem isn't really speed.

 

Anyway, thanks for trying. :)

 

 

As for DIMM compatibility... maybe. The sticks run just as well with two if I mix sticks from either pair. I realise running four sticks means narrower margins, but the fact that the error is always in the same range regardless of the order of the sticks seems to speak to it being something else.

 

The CPU and motherboard were changed in an existing system, so trying the memory out in the old one again isn't really practically feasible, but yes, I have tested the sticks in the recommended two-stick dual channel positions in a variety of combinations. They all work. :)

Edited December 9, 2020 by Ahnion