XMP Profile enabled causes errors in memtest

Hi,

My system has been a bit unstable so i decided to perform memtest to see if the issue was ram related. I have since discovered than my ram is not stable when set to XMP Profile 1 - which is dissapointing.

I have since run memtest for 8 passes (12 hours of testing) with no XMP profile selected at the standard 2133Mhz and it passes no problem.

Can some tell me if there are some additional settings i need to change to get the ram stable? My understanding was the XMP Profile was meant to mean it would be stable on that profile and change all the settings needed for you?

I have tried increasing VCCIO to 1.25v and System Agent Voltage to 1.3v and this didn't fix it.

By the way - I have no CPU overclock and my cooling is average with temps hitting 80c when gaming. So i dont want to OC the CPU at the moment.

***Any help on how to get the ram stable at 4000Mhz on XMP Profile 1 would be greatly appreciated?***

PC Spec below:

CPU: Intel i9 9900K

GPU: nVidia 2080Ti Founders Edition

RAM: Corsair Vengeance RGB PRO DDR4 32GB 4000MHz CL19 CMW32GX4M4K4000C19

Motherboard: Gigabyte Z390 M GAMING

Storage: Samsung 970 PRO M.2 1TB + 860 PRO SSD 4TB

You may be reaching a bit high on memory frequency for that motherboard and 4x8GB@4000 is not a guaranteed target. The XMP 19-23-45 timings are fairly loose at they are. I would up the voltage to 1.375 and test again. If it is still kicking out errors, try 1.40v and move up in small increments.

SA and IO voltage are a bit trickier. More isn't always better for actually stability, although that may not be part of the current issues. I would leave them at 1.20-1.25 and work on the DRAM voltage only. At least you can boot. That means you may not be far off your target.

You may be reaching a bit high on memory frequency for that motherboard and 4x8GB@4000 is not a guaranteed target. The XMP 19-23-45 timings are fairly loose at they are. I would up the voltage to 1.375 and test again. If it is still kicking out errors, try 1.40v and move up in small increments.

 

SA and IO voltage are a bit trickier. More isn't always better for actually stability, although that may not be part of the current issues. I would leave them at 1.20-1.25 and work on the DRAM voltage only. At least you can boot. That means you may not be far off your target.

Hi, thanks for the helpful reply. I appreciate it.

I suspect you might be right. Seems my ram is not on the QVL list for the motherboard either, and not many even are on the list once you get over 3600Mhz. There are only 5 on the list for 4000Mhz and only 1 for 4133Mhz!

I will try the suggestion about increasing the DRAM voltage and see if that works. I am not sure if i want to run over 1.35v 24/7 so I may just have to live to 3600Mhz and try and go for tighter timings. I tested 3600Mhz 16-18-18-36 last night and that passed memtest86!

As you said i guess XMP is not a guarentee as there are lots of other factors at play which ram manufactures cant possibily account for. Just out of curiosity, why does the mother board have such a large effect on ram stability? CPU i can understand...

Not really my area of expertise, but the motherboard circuit traces are critical and memory does not react well to voltage/current fluctuations. More expensive does not always mean better, but often it does mean thicker layers and thus hopefully better conductivity.

You should be able to see RAM temps in iCUE or other monitoring apps (if not using iCUE). I am pretty impressed with my RGB Dominator heat management. In a case without a lot of direct airflow, they still top out at 40-42C at 1.40v in the middle of Summer. This is barely more than running them at 2666 and 1.20v. The V-Pro does not have exactly the same heat spreader, but I would be really surprised is temperature is a factor at anything less than 1.45v.

I don’t put a lot of weight into the QVL. It really means not tested rather than not compatible. Nevertheless board manufacturers often use it as a sword for passing on blame when things don’t work. Unfortunately, there are a lot of other factors involved. I am using a Formula XI and am just now learning the RGB Dominator 3600C16 kit I thought was stable at 4000 is really a cpu-memory nightmare with very erratic performance. That then continues right down the frequency scale. Meanwhile the 4 year old Dominator Platinum 4x4 kit made for my X99 years ago, before Z390 even existed, will crank right up to 4000 with consistent and smooth results. There is a lot of frustration in memory overclocking.

Whatever you decide, know the performance differences between 3600 and 4000 are quite small. You might see +4000 on read/write in an AIDA bandwidth test, but if that mattered I would still use a quad channel X99/299 board that shatters those marks at 2133. The best I ever got was about 3 ns better for AIDA latency and that has no real world value except to OCD memory clockers. My advice is to stay will known stable timings. There are a lot of Z390 boards that get into trouble past 3600, especially with 4 modules or 16GB modules. Subtle, non-terminal RAM errors are even worse than crashing types because you don’t see them and wind up living with sub-par performance.

Not really my area of expertise, but the motherboard circuit traces are critical and memory does not react well to voltage/current fluctuations. More expensive does not always mean better, but often it does mean thicker layers and thus hopefully better conductivity.

 

You should be able to see RAM temps in iCUE or other monitoring apps (if not using iCUE). I am pretty impressed with my RGB Dominator heat management. In a case without a lot of direct airflow, they still top out at 40-42C at 1.40v in the middle of Summer. This is barely more than running them at 2666 and 1.20v. The V-Pro does not have exactly the same heat spreader, but I would be really surprised is temperature is a factor at anything less than 1.45v.

 

I don’t put a lot of weight into the QVL. It really means not tested rather than not compatible. Nevertheless board manufacturers often use it as a sword for passing on blame when things don’t work. Unfortunately, there are a lot of other factors involved. I am using a Formula XI and am just now learning the RGB Dominator 3600C16 kit I thought was stable at 4000 is really a cpu-memory nightmare with very erratic performance. That then continues right down the frequency scale. Meanwhile the 4 year old Dominator Platinum 4x4 kit made for my X99 years ago, before Z390 even existed, will crank right up to 4000 with consistent and smooth results. There is a lot of frustration in memory overclocking.

 

Whatever you decide, know the performance differences between 3600 and 4000 are quite small. You might see +4000 on read/write in an AIDA bandwidth test, but if that mattered I would still use a quad channel X99/299 board that shatters those marks at 2133. The best I ever got was about 3 ns better for AIDA latency and that has no real world value except to OCD memory clockers. My advice is to stay will known stable timings. There are a lot of Z390 boards that get into trouble past 3600, especially with 4 modules or 16GB modules. Subtle, non-terminal RAM errors are even worse than crashing types because you don’t see them and wind up living with sub-par performance.

I think thats a very good summary, thanks for explaining all of that. The last point you make is key, stability has to be number 1, i have been having crashes for a long time and it's because my ram has been unstable ever since i installed windows about 6 months ago and swtiched on the XMP profile. Everything i have since installed has probably had some errors baked in due to the ram instability. As you say, thats really incidious to the system and litters everything with errors and these build up more and more over time.

Tomorrow i am taking off all ram overclocks and installing windows and all drivers again from scratch and then only using the 3600Mhz profile thats passed a full memtest, (or only use other profiles i have proved are 100% stable in memtest). I think i just took XMP as a gaurentee and when i saw 4000Mhz displayed in windows assumed that would be stable and never thought to test it.

My mother board is not anything super high end, because i am using mATX and there are only a small choice in this space. Definately very ambitious of me buying such high frequencey ram even though i saw the motherboard spec saying it could suppprt up to 4133, but when you look at the QVL list they only have 1 set of ram listed as tested at that speed! Probably just to be able to market the board as "speeds up to 4133Mhz". I will be happy if i can push the timings down very low on 3600Mhz, as you have said, real world performance is not massive when going up to the super high frequencies. I have seen lots of charts saying ram speed is more or less irrelevant when gaming at 4k as its all VRAM on the GPU at that res.