2x Quad-Channel Kit Compatibility?

I have an older x99 build that currently uses: CMD16GX4M4A2666C16 (16GB Corsair Dominator 4x4GB Quad-Channel 2666MHz C16) on an ASUS Deluxe X-99 MoBo (usually overclocked to 2800MHz on 1.35v). The X-99 MoBo has 8 RAM slots. I want to upgrade to 32GB as I am finding it often above 80% RAM usage. The CMD16GX4M4A2666C16 is no longer being manufactured, but I was able to find a few for sale online.

 

In theory, I should have no problem running 8x4GB in quad-channel. Am I missing anything?

Hey Mordred,

 

Thank you so much for the reply. I went out and bought the second quad channel kit. Installed it. And of course, there is a problem.

 

The BIOS/CPU (and subsequently the Windows 10) isn't detecting 2 of the DIMMs as being occupied.

This is interesting because it is DIMM_C1 and DIMM_C2, separate channels. As in its not the new RAM. It is one stick of the old RAM (which used to work in that DIMM) and one stick of the new RAM.

 

Making things more interesting, HWInfo and CPU-Z both detect all 8 DIMMs and 32GB:

 

Every explanation I can come up with is shot down by the fact DIMM_C2 used to work and the fact I did not touch it or the fact that all 8 DIMMs are showing as correctly occupied in HWInfo and CPU-Z.

Fixed. What a fascinating problem.

 

The solution for anyone else that might have the same issue: Set DRAM Voltage to AT LEAST 1.20

 

On my motherboard in particular (ASUS X99-Deluxe) I noticed, with the DRAM voltage set to auto, DIMM_A/B (left side) would sit pretty steady at 1.200 or slightly higher. However, the DIMM_C/D (right side) would bounce sometimes hitting 1.19 so I manually set the voltage to 1.20 (from Auto) and ta-da. Now have all 8 DIMMs, 32GBs, running steadily in XMP 1.

 

Edit: Yes, interestingly enough, DIMM_C1 and DIMM_C2 are the far right DIMMs.

It was not uncommon for those first Asus X99 boards to register different voltages between the AB and CD sides. Whether the differences are real or not is another matter. Also know the Deluxe/Pro/A were always a bit finicky about physically inserting the modules. I can't tell you the number of times I took them out to reconfigure some piece of hardware, then spent 20 minutes continuously plugging and replugging to get all the sticks recognized. It made me loathe to ever take them out. That is something that almost never happened later on the second gen. X99 boards or now on my Z370 running the same kit.

 

If you are going back up to 2800, you'll need more than 1.20 anyway so you can always shade another hundredth or two more to be on the safe side.

I recall having trouble with inserting the modules some years ago when the build was in progress. This was something I thoroughly and repeatedly checked throughout the trouble shooting process. However, I mostly ruled out this being a possible point of malfunction when I realized CPU-Z and HWInfo were correctly detecting all modules.

 

I am curious, since you mentioned it, why the second gen and Z370 boards evaded this problem. (My new build is on an EVGA X299 Dark)

 

I would like to work my way back up to 2800 @ 1.35V, or probably 1.37V to be safe. But I don't think the benefits outweigh the increased power, thermals, and instability with 8 modules.

 

I would, however, like to get my CPU clock back up to 4+GHz (its running @ 3.6GHz now) but hesitate because without the faster RAM speed I'm not sure if that becomes entirely beneficial.

 

Thoughts?

Definitely do the CPU clock first. 2800 vs 2666 is not an overly advantageous position. You can usually get away with lower voltage 1.20-1.25 at 2666 while getting 99% of the performance. I also just don't like 2800 as a speed and had a hard time with it on that board. All the X99s should be able to run 2666 upside down, sideways, and with poor timings. This seems like the better choice for professional work.

 

Why did the boards get better? I don't know for sure, but X99 was the first DDR4 platform and presumably motherboard manufactures got better at things. X299 certainly has better capacity handling than X99 and the process continued to improve. I just know from experience it was enough of hassle to stop me from taking the modules out on the X99, then only needing to re-seat once for the next two platforms with the same sticks and dozens of remounts. Come to think of it, my very first X99 went back to the vendor for this issue. The replacement was better, but as described above. It sounds like you are sorted now either way and adding an extra 0.01 to your DRAM voltage will be meaningless from a temperature standpoint.

Update: I returned the second kit today. Just couldn’t get the system stable in any configuration. To be fair, I am sure the limitation was on the motherboard.

 

I could get the system stable and running wonderfully all day. However, turning the system off for the night and cold booting it the next day resulted in BSOD, clear CMOS and doing it all again. After three days of this--wasting over an hour each time getting the system stable again--I gave up.

 

Edit: for example, right before I returned the memory kit I ran memtest86... failed in the first 3 mins. On just the original memory, memtest86 ran for over an hour with no errors.

 

I guess this is just another testament to how finicky mixing RAM can be.

Yes, I am afraid that is true. It is possible you could have stabilized them eventually, but it probably would require dozens of hours of testing and tweaking the sub-timings. I love to play around and over-optimize everything on my gear, but not memory. It requires a different level of patience I just don't have the time for. As expensive as DRAM is right now, a week (or two) spent trying to solve that kind of thing is worse than paying for 8x8 or 4x16 kit - if I really needed it.