Does CPU core voltage affect ram system stability?

This is an odd one I've been dealing with since building my current main system: Asus VIII Hero, 6700k, currently G.SKill Ripjaws V 3200C16 2x8GB.

I do a lot of prime finding projects, which may be compared to running Prime95 large FFT option. It never gives errors if I do tasks equivalent to small FFT option, which would not significantly involve the ram system. In the early days I had all sorts of compute stability problems, but between bios updates and changing the ram, it seemed to have gone away. Before that, I was running G.Skill RJ4 3333 and the best I got that was one detectable error in 3 months, which is still higher than zero errors ever on other systems. Note these tasks are significantly affected by ram speed, hence my hesitation to drop it if possible.

Recently, I got another detected error, and I'm kinda wondering what to try next. The giving up solution would just be underclock the CPU and ram a bit and call it a day. Instead, I have decided to increase Vcore slightly, from 1.250 set, to 1.275 set. My reasoning for this is that my other two Skylake-K systems I've found also need 1.25V set for stability at 4.2 GHz, just like this one. The difference is... the Asus mobo seems to have higher voltage droop under load. I've not made use of load line stuff, and under load it was dropping well below 1.200. Again, I've not had any stability problems with small tasks, only bigger ones. Does CPU Vcore in any way affect stability of the memory system? Note I say memory system, as it could include everything from the IMC to the ram itself.

I've also updated the mobo bios as I was a few versions behind, and at least two of the ones I missed say "improved system stability".

For indication, I was running the type of work which got an error for about 4 days before it happened. Historically we're dealing with very low error rates, from about one a week to one a quarter. I still believe it should be exactly zero.

In theory if you are not using ECC RAM, you can't get to 0 for super long periods of time.

All it takes is a few stray cosmic rays. In my defense work years, we used to use "hardened" memory and CPU components to prevent this.

You could try more vcore - as long as you stay under 1.4 V you won't damage anything...just extra heat.

You could also try giving the memory more voltage as well.

Woomack can pipe in for more on memory.

Does CPU Vcore in any way affect stability of the memory system? Note I say memory system, as it could include everything from the IMC to the ram itself.

Click to expand...

It should not, no. There are different voltage domains for vCore and vDIMM and the other voltages that affect the memory/IMC. Cache is tied to the vCore voltage however... but not sure if you call the CPUs cache part of the memory subsystem.

The voltages on this platform that affect memory the most are the System Agent (SA) and I/O voltages (for the IMC) and of course the voltage for the sticks themselves.

If you are having a lot of vdroop, perhaps its the CPU borking things and not the memory (it does afterall have to process that data). Definitely get that stable(r), and then see what's shaking with the memory.

Since your vCore is dropping so low under load I would most definitely give that a shot long before I would dismiss it as an option. My Asus board with a Load Line of level 6 prvents any drop in voltage under load and stays the exact same under load. Level 7 will increase it under load and 5 will still cause it to lower under load. And being as vCore in the case of Skylake controls Cache as well as Core Ratio it is most definitely something I would adjust. I can't sit here and act like I know wth your error was about but I kind of associate Cache with "Memory" per se.


Nice E.D. Lol

Good points by all. My reasoning for not suspecting the CPU directly is that small FFT tasks remain error free. Only large FFT tasks may show errors. For the purposes of these tasks, I'll exclude the L3 cache from consideration, as even the small tasks will need to use that. Thinking more, maybe there is more droop under large tasks than small ones, which isn't something I looked at specifically so far. I will try that when home. I guess also using load line might be more elegant than setting a blanket higher voltage so can try that also.

I think my next main system will be a server platform with ECC, and I'm already eyeing up some dual Xeon systems... I'm more tolerant of dedicated crunchers being unstable since all work is double checked anyway, but I do like my main system to be stable.