RAM no longer boots with XMP profile after upgrading video card

[ExpedientFalcon]

Prior, I was running a Geforce 660GTX, and my system would run fine on the XMP profile at 2666Mhz/1.35v. (I say mostly because there were very occasional system lockups, but I could never verify if these were due to the memory--Memtest reported 0 errors after multiple passes.) However, after upgrading to a Geforce RTX 2070, the system refuses to boot at the XMP profile. It will boot cycle to the BIOS screen a couple of times, then auto-revert to the JEDEC setting of 2400Mhz/1.2v, which boots fine. However, another weird behavior which I just noticed is that the graphics card appears to be reserving 8GB of system memory--the Windows task manager shows only 8GB usable, with the other 8GB reserved by hardware. This seems strange, since the graphics card has 8GB of its own RAM, but I couldn't find a setting in the BIOS to alter this behavior.

Has anyone encountered anything like this before? Below are the system specs:

AMD Ryzen 2700X
Gigabyte X470 Aorus Ultra Gaming
2x8GB Kingston HyperX HX426C13PB3K2
EVGA Geforce RTX 2070 XC Ultra Gaming 8GB
Seasonic Focus 750W Gold PSU

I also verified that this RAM is on the mobo compatibility list.

 

[Zerileous]

This is my best guess: Ryzen and specifically Pinnacle Ridge use a System on Chip (or SOC). This will run at the speed of the memory and contains the Northbridge and Southbridge (in otherwords, IO, PCIe, memory controler, etc). When you enable the XMP profile of your memory you overclock the entire SOC. Prior to the upgrade this overclock was (mostly?) stable, however the added demands (probably in PCIe bandwidth) of the more powerful card made this overclock of the SOC unstable.

The solution, if my theory is correct, is to give more voltage to the SOC. This may affect your CPU temps, so good cooling is important. According to the Ryzen Tool Kit thread, 1.15v is the maximum recomended setting for normal use and lognevity. I'm running my 2600X at 1.0v right now with memory at 3200mhz and a vega 64. You could try setting it to 1.0v and see if that works, if not, increase by 0.05v until either stable or you reach 1.15v (although I seriously doubt you'd need this much.) If you have trouble finding this setting, please post pictures of your BIOS and we can help.

I also used a program called DRAM calculator and entered a nauseating number of sub-timings into my bios because I was having some instability with the XMP profile. Since then no issues. So you could try that as well.

Also wouldn't be a bad idea to post pictures from CPU-z and GPU-z so members can have a better idea of what all your hardware is doing.

 

[ExpedientFalcon]

Thanks! It almost runs at XMP (only increases the CAS Latency by 1) with +.1 on the SOC voltage (so it's running at .975v), although I'm not certain if it's stable--I got it to boot once before at stock SOC voltage, but it froze after a minute. But, the system is now using all 16 GB of RAM, so something has improved.

 

[ExpedientFalcon]

Update: No good, crashed during Prime95. Now I can't get it to boot back up at 2666Mhz, even if I push the SOC voltage as high as this mobo will take it (which is about 1.15v). So maybe I just got lucky for a bit on that one try.

 

[Zerileous]

Is the CPU overclocked or just using its stock boost settings? I can't tell. How are you cooling the CPU? You could try 14,16,16. I read somewhere that Ryzen likes even timings. I would also try updating the BIOS, maybe it is being thrown off by the newer card. Other than that I would personally go through the DRAM calculator I linked above.

My memory wasn't on my QVL (asrock had a very short QVL and G.Skill doesn't test much trident on AMD systems), but knowing it was B.die which plays well with ryzen, I went with it. I still had trouble just running the XMP out of the box. Even though you're not going nearly as fast, it's the only idea I have at this point, and your memory is on QVL, if the bios flash doesn't work the subliming calculator is the only idea I have at this point.

 

[ExpedientFalcon]

It's not manually overclocked, but the setting that lets the CPU overclock itself is enabled. It's on stock cooling but temps are very healthy, peaking around 60c. BIOS is on latest version. I'll give the even timings a shot. I tried the DRAM calculator but it was giving me something like 10-12-12 for safe settings (maybe I was using it wrong, though, so I'll try it again and see if I can get better numbers from it).

 

[Johan45]

Have you tried the XMP for 2400?
It might work better and then maybe speed it up after

 

[ExpedientFalcon]

XMP for 2400 was not stable

However, after going back and reading through the instructions for the DRAM Calculator and figuring out what I was looking at, and why there is even a DRAM Calculator for Ryzen (because the XMP settings are designed for Intel CPUs), I went back and applied those in my BIOS and the system does appear stable at 2666-14-16-16-36 now. Going to run some Prime95 and hopefully come back with positive results. It's a bit odd that I had to do this, since the "XMP" box was checked as supported for this RAM on this motherboard, but at least it seems better now. Thanks for the help!

 

[ExpedientFalcon]

Just kidding, crashed after 5 minutes in Prime95. Feels bad man. I'm going through the recommendations in https://www.overclock.net/forum/27694284-post3363.html but unsure if any of these will help--it feels like I shouldn't need to go this far into advanced settings to get basic system stability.

 

[Zerileous]

Just to clarify, are you 100% stable running the JDEC default? Even the XMP is not that fast, lets rule out other causes of instability. Also maybe try V2 in the calculator, I think it uses more relaxed timings.

I know what you mean, feeling like it shouldn't be this difficult. The way memory is marketed makes it seem like an advertised overclock is a specification. I find it to be really misleading, personally, but that is how the market is. Anything beyond JDEC is overclocking both the memory and the CPU. Still a lot of variables in the silicone lottery to account for, but you should be able to achieve a stable setting at 2666.

Edit, after looking at the link you posted, check out the Alt 1 and Alt 2 settings for procODT and RRT.

 

[ExpedientFalcon]

No, Windows would only use 8GB out of the 16GB with the JEDEC default. Now, after swapping the memory sticks between different slots to see if that made any difference, even if I put them back in the original slots, I'm getting frequent crashes and freezes at JEDEC settings. Before switching out the GPU a couple of weeks ago, I would get occasional freezes at 2666Mhz, but it was only under very specific workloads, so it was not completely stable then either, but it has gotten much worse since upgrading the GPU (maybe coincidence, maybe related).

Is this the point where I should try a different set of RAM?

 

[EarthDog]

The way memory is marketed makes it seem like an advertised overclock is a specification. I find it to be really misleading, personally, but that is how the market is.
Anything beyond JDEC is overclocking both the memory and the CPU.

For the CPU IMC, it is overclocking. For the memory, it's literally rated to run that way out of the box...its stamped all over it is rated to run at that speed. ALL memory IC's are derived from the same speed silicon (2133/2400), indeed, it is then binned and stamped with a speed its rated for. Binning like that, to me, isnt overclocking. If that is overclocking, all CPUs are overclocked? If its rated at X speed for its specification, it isnt overclocked to me...

Due to board and IMC differences, sometimes it doesn't work out. For AMD, Ryzen especially, they are a lot more finicky than Intel is with memory.

Edit: isnt JEDEC on X299 overclocking? Base there is 2666... if ram ics are 2133/2400 base... JEDEC is 2666 in that platform.

Sorry, this is sort of an aside, lol....I digress.

 

[Zerileous]

Fair enough point EarthDog. I even bought some Mushkin Black Label DDR back in the day because the warranty covered overvolting, so I guess in a way I supported the practice in its infancy

How long did you have the rest of the system before upgrading your GPU? Can you reproduce the error where only one stick is detected? I would go into your bios and restore default settings if you haven't already done so. If you still have access to the old card, I would also re-install the old card if the system isn't stable at default bios settings. If it's unstable with both GPUs then probably the memory is at fault. If you have access to a different set of ram, definitely try that as well. You can also try booting one stick at a time and testing stability. This won't give you particularly useful performance, but it will definitely showcase a bad stick which would easily support a warranty claim.

 

[ExpedientFalcon]

The core of the system (CPU/mobo/RAM) is about 8 months old, and the old GPU was not upgraded from the previous build before that, so it was 4-5 years old. I've tried swapping out one stick at a time, or using one stick in different slots, with no improvement. Also tried swapping back to the old GPU, and no good. Memtest86 crashes after about 1 minute, even at JEDEC settings. I ordered these off this recommendation that Samsung-based chips work well with Ryzen, so we'll see tomorrow if the new RAM helps.

 

[Zerileous]

Wow have memory prices dropped.
I'm really not certain that the sticks you ordered are composed of Samsung chips, but that might not matter as much as people think. It gets really confusing with all the different product numbers. The safest bet would be to order something listed on G.Skill's RAM configurator for your motherboard. That's not to say that the ones you selected wont work, looks like a few Ryzen builders have left reviews in favor of them. I also see sticks with similar timings listed in both the QVL and the link above. I'm just going off of memory here, but one of my co-workers is using red Ripjaws on a x470 Aorus (not sure which one) with the 2700x as well, and I think its the 3200 16-18-18, but it could also be a 3400, but he hasn't had any problems with them.

Can you clarify if you restored bios defaults? Also I would consider flashing the bios, even if your current version is up to date, something may have been corrupted.

 

[ExpedientFalcon]

Yes, I've restored the BIOS defaults. I'm actually afraid to flash the BIOS at this point, given that the system has frozen at the BIOS screen and I don't want to brick it.

Here is where I got the info that those Ripjaws are using Samsung chips. Thanks for the link to the configurator. I may pick these up instead since they're on the configurator list, and the 3200Mhz variant seems to be running Samsung B-dies. Also, it's insane what the price difference is between 16 and 14 CAS. All the 14 CAS sticks are over $200.

 

[Zerileous]

Just to make sure, use the clear CMOS jumper on your motherboard.

I'm still concerned that it could be a motherboard or CPU problem. Memtest crashing doesn't necessarily mean memory is at fault, if it doesn't display errors prior to the crash. I'm having a hard time figuring out what's going on here. It seems hard to believe that two sticks would die at the same time. It also seems hard to believe that a compatibility issue could evolve into complete instability at previously stable settings in two days. On the other hand, the failure started as instability at XMP and progressed to instability at JDEC settings, which could point to RAM. It could also be some setting in the bios that didn't get reset properly (hopefully the CMOS jumper would settle this).

I didn't realize bios was unstable, in that case I don't blame you for wanting to hold back. I thought I searched that page for the sticks that you ordered, but I must have missed it. I doubt the sticks you linked are running B die, probably E or D die. The CL 14 sticks are B die, and that is why they cost so much (so far only b die can run 3200 14-14-14). I paid less than $200 for my kit over black Friday, but I still paid a (likely unnecessary) premium.

I honestly don't know if there is a difference in sub timings and stuff between the ripjaws V and sniper X, or just what systems they get tested on. Either way you might want to hold off on ordering the memory until we're sure that's what is wrong.

 

[Johan45]

I understand your misgivings with the RAM, they shouldn't crash at JEDEC in memtest. In general, you'll have better luck with RAM that's intended for the AM4 platform like the FlareX and if you're not benchmarking etc... you'll be fine with Hynix based RAM which anything at CL16-18-18 and 3200 most likely is. I also had good luck with the sniperX 3400 kit that I tested https://www.newegg.com/Product/Product.aspx?Item=N82E16820232727 It's Samsung based but a slight variant of the cherished "B" die variety so not quite as pricey.

Aside from changing graphics cards have you updated the BIOS lately? I notice on their site you have the latest BIOS but have you updated the chipset driver like it recommends?

 

[wingman99]

For the CPU IMC, it is overclocking. For the memory, it's literally rated to run that way out of the box...its stamped all over it is rated to run at that speed. ALL memory IC's are derived from the same speed silicon (2133/2400), indeed, it is then binned and stamped with a speed its rated for. Binning like that, to me, isnt overclocking. If that is overclocking, all CPUs are overclocked? If its rated at X speed for its specification, it isnt overclocked to me...

Due to board and IMC differences, sometimes it doesn't work out. For AMD, Ryzen especially, they are a lot more finicky than Intel is with memory.

Edit: isnt JEDEC on X299 overclocking? Base there is 2666... if ram ics are 2133/2400 base... JEDEC is 2666 in that platform.

Sorry, this is sort of an aside, lol....I digress.

CPU's are binned at the factory of the processor company. Memory company's use the same high quality part numbers from the IC factory then the memory stick manufacturers raise the JDEC voltage and bin the memory with overclocking testing them in benches. I feel Intel's and AMD testing has more tolerance and more accuracy. Memory company don't build tolerance into there memory overclocking from the questions I asked.

 

[EarthDog]

Lol, I'll never call using XMP overclocking the ram sticks.

That bit of information (what the ics really are) yields so little use to people, it's not even worth mentioning IMO. What value does it bring to 99.9999% of people? The sticks are rated to and will run at their rated speeds. They are tested to do so and deemed to run at that speed, packaged and spec'd as such.