ram unstable at rated speeds

[Celach]

Yesterday i finished building my pc but it seems the ram i bought is unstable when running at rated speeds. I bought everything on the US from amazon but now im back home in south america so i dont really know how easy or costly would it be to ask for a replacement. I have enabled xmp default profile and ive changed the settings manually in the BIOS but both seem to be causing game crashes.

Should i try to ask for a replacement or do i just settle for a lower speed? my ram is at 2133 mhz atm, while the rated speeds are 3000. I'd really like to use my ram at 3000 mhz since, from what ive read, higher speeds reduce stuttering a bit, but i dont really know if i should try increasing the ram voltage to have it @3000 mhz cause i dont know how badly it would affect the ram. Any guidance is appreciated, thanks

this is my pcpp: https://pcpartpicker.com/list/fQRJmq
p.s i havent tried middle ground speeds nor increasing the voltage to check if it gets stable.

 

[boju]

When xmp was set did it set dram voltage to 1.35v? Not 1.2v?

Do you have the latest bios revision?

 

[Darkbreeze]

In almost ALL cases, running memory at speeds much higher than the default SPD will usually require some kind of small CPU overclock at the very least. Are you running your CPU at the default settings or is it overclocked?

I'd bet a small overclock plus a minor bump in DRAM voltage will allow those to run at their rated speed of 3000mhz. In my experience it's very hard to get memory to run or even POST at 3000mhz or higher without overclocking the CPU at the least. Often it requires that plus a bump in DRAM voltage but I'd try to just bump the CPU up a bit first. If that doesn't do it then increase the DRAM voltage a bit or manually set the DRAM voltage to whatever the XMP voltage spec states, and then if that doesn't work try increasing voltage to the DRAM by pressing the + key one time to incrementally increase voltage by one increment. Every board seems to be different on what increments they allow you to adjust voltage so telling you to increase DRAM voltage by .010 or .020 may not be accurate for your board. It may be more or less, incrementally, than boards I'm more familiar with.

Also, as Boju mentioned, make sure you have the latest bios installed. Personally, I'd only update to versions that are not newer than December of 2017 so that you don't get hit with the performance penalty from the microcode updates for Spectre and Meltdown, but that's up to you. Your board may already have a newer bios than that or you may feel more comfortable having those updates in place anyhow.

 

[Karadjgne]

If you look at the ram speeds supported by the mobo, in every case of ram at higher speeds than cpu memory controller default (2133MHz for skylake) will all have an (OC) next to the ram number. This means 1 of 2 things, possibly both. First is simply to get rated speeds will require the ram be overclocked by XMP profile or manual settings. Second is the memory controller in the cpu might need a small boost in performance to be able to handle the higher speed ram at a stable speed. This means OC of the cpu. Part of the reason you will not see OC ram speeds on a non overclockable H or B rated board.

Your stuff isn't broke. It doesn't require return or RMA. It doesn't need fixing. What it does need is to be tailored to your particular use by adjustments in the bios of either cpu OC, XMP, combination or other ram voltage capacities

 

[Celach]

Thanks for the replies, I thought my system was supposed to handle that speed at once without any more tweaking other than setting up xmp. What ive done at the moment is update to the newest bios and manually set the xmp profile myself, with rated voltage, timings and speed. Though i havent started overclocking my cpu yet. Ill start testing now your suggestions, thanks.

One thing though, if i end up needing to add more dram voltage, whats a safe range i should go for?

 

[boju]

Shoot for 1.35v dram

 

[Darkbreeze]

DDR4 can go all the way up to 1.5v on some XMP values for high speed modules. I would try to go no higher than 1.38v max. Common sense and experience say that taking your CPU to about 4Ghz and increasing the DRAM voltage, if the CPU overclock by itself doesn't cure the problem, to about 1.36v, should probably resolve your issues.

If your bios has the option, I'd also set your training voltage to the same voltage as what you intend to run the modules at. I've had issues leaving the training voltage (Might have different nomenclature on different brands of board) on auto occasionally. The board will train the secondary and tertiary memory timings differently in some cases based on different training voltages.

 

[Karadjgne]

I start with what XMP sets and then add 0.005v. So if XMP sets 1.35v, my next bump would be 1.3505v. My MSI mobo has a drop down menu for this, preset voltages to choose from, so there's no manual guessing, I just move to the next voltage down. Can take a while and multiple reboots to get final stable results, but doing it in small increments will keep voltages at minimum requirements and not dumping a bunch of unnecessary voltages through the ram.

 

[Darkbreeze]

Many boards don't support those increments. And then, many do. So it just depends on what your board supports which is exactly why I said earlier to use the + key, so you can see what increments your board supports for adjusting up and down.

My Z170X-Gaming 5, for example, does not allow adjustments in such small increments. You can only make changes to DRAM voltage in increments of .020v. So, 1.34, 1.36, 1.38 and so on. Being as the Z370-A is a somewhat mid tier board it may not have exceptionally granular options either, or, being an ASUS board, it might. Hard to say. They change so many things from board to board and gen to gen. If you haven't worked on THAT or ANY specific board, it's very hard to say and the user manuals are generally useless in this regard, as they never offer that kind of information.

 

[Karadjgne]

Well yes, of course, my board accepts increments of 0.005v, so I just choose the next down in the list. If op's board is 0.02v then he'd move to that next one down. Just iterating the point that there's no point setting 1.5v right from the start and leaving it there.

 

[Celach]

I currently have my cpu @4.8 ghz and ram at 3000 mhz @1.36v but games still crash, is there anything else i should do?

 

[Karadjgne]

Check windows event viewer. Look for the critical errors, it might have a clue where to start looking

 

[boju]

Was the system running fine with memory at 2133?

 

[Celach]

Was the system running fine with memory at 2133?

yes whenever i disable xmp games stops crashing

 

[boju]

Got these settings from here https://rog.asus.com/forum/showthread.php?98925-Prime-Z370-A-memory-instability-problems

Try these as manual overclock. What timings does your memory use? Try the timings xmp uses and then raise the two middle numbers by one decimal. Set command rate to 2T.

Im guessing your memory is 16.16.16.36?

Example; adjust bios 3000MHz CL16-16-16-36-2T or CL16-17-17-36-2T

And;

DRAM : 1.3530
VCCSA: 1.1500
VCCIO : 1.1375

 

[Darkbreeze]

@Karadjgne, nobody said anything about setting the voltage at 1.5v from the start. That's just being asinine.



@Celach. You're trying to jump forward way too quickly. You need to incrementally overclock the CPU first. Start with something closer to your stock speed. No way you can jump to 4.8Ghz on your CPU, immediately, and know it's stable, and then jump right to the DRAM. It doesn't work that way. You need to get the CPU stable, which takes hours of testing, before you even think about changing the DRAM from the default speed.

What is your CPU voltage at 4.8Ghz?

What is your load line calibration set at?

Is EIST enabled or disabled?

Is Turbo boost enabled or disabled?

Have you run Realbench for 8hrs?

Have you run Prime95 version 26.6 small FFT or Blend modes? For how long?

Then, when you start making changes to your memory speed, voltage and possibly timings, you need to run several passes of Memtest86 every time you change something, followed by 8hrs of Prime95 version 26.6 Blend mode, preferably with custom settings of min 512k fft and max 4096k fft using approximately 75% memory allocation.

Until you are ready to do all these things, you might as well just run everything at the stock speed. If you are ready to do these necessary and highly crucial steps, let us know, otherwise, all these other stuff and commentary is just drivel.

 

[Celach]

@Karadjgne, nobody said anything about setting the voltage at 1.5v from the start. That's just being asinine.



@Celach. You're trying to jump forward way too quickly. You need to incrementally overclock the CPU first. Start with something closer to your stock speed. No way you can jump to 4.8Ghz on your CPU, immediately, and know it's stable, and then jump right to the DRAM. It doesn't work that way. You need to get the CPU stable, which takes hours of testing, before you even think about changing the DRAM from the default speed.

What is your CPU voltage at 4.8Ghz?

What is your load line calibration set at?

Is EIST enabled or disabled?

Is Turbo boost enabled or disabled?

Have you run Realbench for 8hrs?

Have you run Prime95 version 26.6 small FFT or Blend modes? For how long?

Then, when you start making changes to your memory speed, voltage and possibly timings, you need to run several passes of Memtest86 every time you change something, followed by 8hrs of Prime95 version 26.6 Blend mode, preferably with custom settings of min 512k fft and max 4096k fft using approximately 75% memory allocation.

Until you are ready to do all these things, you might as well just run everything at the stock speed. If you are ready to do these necessary and highly crucial steps, let us know, otherwise, all these other stuff and commentary is just drivel.

Im sorry, i get what you mean. I tested the overclock at 4.8 @ 1.29v (lower voltages were unstable almost right away) for about half an hour with prime95 and thought that was okay for it to be stable. I also enabled xmp at the same time i was overclocking, so i guess that was wrong too.

I'll start overclocking calmly now and run over night tests before i move on to the ram oc. sorry again if im a bit impacient, i appreciate your help.

As for the other settings you mentioned i didnt really touch anything but avx offset and core clock/voltage, ill research on those you mentioned now though, thanks again

 

[Darkbreeze]

No possible way a 4.8Ghz overclock is stable on any modern Intel platform with only 1.29v core voltage setting. That might be what voltage drops to when you look at the monitoring, but if that is the actual setting you can bet that under a load the voltage is dropping considerably below that unless you have load line calibration set at the highest levels.

I would probably start at about 4.3Ghz with like a 1.32v setting. Leave the memory at default SPD settings for now. Run Prime95 version 26.6, no other version, due to the unrealistic AVX instruction load, for about 15 minutes to establish that you are thermally compliant. If you are, then run Realbench for about 4hrs with 8GB memory option selected. Later you will test full 16GB using this and other tests, but right now we only want to see CPU stability and are not concerned with memory stuff.

If you want to know a baseline determination that the memory itself is not faulty, before beginning any other steps, create bootable Memtest86 media and run two full passes with the CPU at default settings after which you can then configure your CPU overclock and begin that testing.

It would be highly advisable that you download all of the following for use while testing and configuring your setup.

Realbench
Prime95 version 26.6
Memtest86
HWinfo
CoreTemp

I would advise against using anything other than HWinfo and CoreTemp to monitor thermal compliance and other sensor values. Too much and too many cases of erroneous readings with other utilities depending on what platform and chipset are being used. These utilities tend to have the broadest and most accurate sensor and chipset compatiblity and are updated pretty religiously, so they are, to my belief, the most trustworthy of the utilities available.

 

[Karadjgne]

I use Real Temp, it's specifically designed for use with Prime95 and Intel chipsets and it runs the same temps as CoreTemp as far as accuracy goes. I also like the fact you can stick it in the Taskbar so it's constantly monitoring temps at a glance, no matter what else is on the screen. (if in windowed mode, of course)

 

[Darkbreeze]

I've actually never used Real Temp, so I can't speak to that but I know for a fact that I personally, and MANY others that I've run into both on this forum and others, have have hit and miss reliability and sometimes just plain bogus readings using HWmonitor, Openhardware monitor, most of the motherboard manufacturers desktop monitoring and overclocking utilities, Windows various monitoring utilities and Speedfan, among a few others.

As far as being designed for use with Prime95, I don't know really how that matters. Either the utility you are using is reporting accurate sensor data or it isn't. There's really no stress test specific factors that would change that that I'm aware of.

 

[Karadjgne]

It's got a button to open p95. I'd also assume since they claim it's written specifically for use with p95 that any of its coding is done in such a way as to provide 0 possible conflicts with p95, something which others like Hwmonitor etc with its much more in depth reporting/monitoring etc may not do. But thats just a guess. As is, realtemp is pretty basic, realtemp GT is for more than quad cores.

But it's still software. And yeah, there's a lot of software I've used and so far relatively few are 100% reliable, even for something as simple as the 12v rail voltage. I've yet to run anything that actually reports my system at 12.18v (measured physically), they all report either 8.2v or 10.2v. Most programs also report 2 sensors, 1 at 255°C, the other at -125°C.

Typical software failures. I've run both CoreTemp and RealTemp on my pc's, both had the same results, I just prefer the simplicity and ease of use with RealTemp. Just does temps of cores, no tmpin0 or other junk to confuse with, just core0-3

 

[Darkbreeze]

CoreTemp. I don't know how it can get any simpler than this to be honest.

 

[Celach]

No possible way a 4.8Ghz overclock is stable on any modern Intel platform with only 1.29v core voltage setting. That might be what voltage drops to when you look at the monitoring, but if that is the actual setting you can bet that under a load the voltage is dropping considerably below that unless you have load line calibration set at the highest levels.

I would probably start at about 4.3Ghz with like a 1.32v setting. Leave the memory at default SPD settings for now. Run Prime95 version 26.6, no other version, due to the unrealistic AVX instruction load, for about 15 minutes to establish that you are thermally compliant. If you are, then run Realbench for about 4hrs with 8GB memory option selected. Later you will test full 16GB using this and other tests, but right now we only want to see CPU stability and are not concerned with memory stuff.

If you want to know a baseline determination that the memory itself is not faulty, before beginning any other steps, create bootable Memtest86 media and run two full passes with the CPU at default settings after which you can then configure your CPU overclock and begin that testing.

It would be highly advisable that you download all of the following for use while testing and configuring your setup.

Realbench
Prime95 version 26.6
Memtest86
HWinfo
CoreTemp

I would advise against using anything other than HWinfo and CoreTemp to monitor thermal compliance and other sensor values. Too much and too many cases of erroneous readings with other utilities depending on what platform and chipset are being used. These utilities tend to have the broadest and most accurate sensor and chipset compatiblity and are updated pretty religiously, so they are, to my belief, the most trustworthy of the utilities available.

Hey, it's probably been way too long but i figured i would update. So I've been running overnight tests and reached a stable overclock of 4.8 ghz @1.34v and level 4 llc (medium level with z370 prime-a), for that i left prime95 small ftt and realbench overnight (6-9 hours). Then i moved into the memory speed but 3000 mhz was never stable (memory86 presented errors in less than 5 minutes from 1.35-1.37v), not even at a low oc like 4.5ghz with low voltage, so i reduced it to 2933 and its stable at that speed so i guess ill leave it there, i dont really think theres anything else i can do, maybe i got unlucky with the 8600k imc (?)

After leaving it at 2933 mhz @ 1.35v and rated timings i ran prime95 blend test, memory86 and realbench with 16gb memory overnight and there were no problems, so i guess im leaving it as it is. Its really close to 3000mhz so its not a big problem anyways i think. Thanks for your help

 

[Darkbreeze]

There are a few things you can do to help improve the memory performance, and then retest if you wish to.

Find the settings for VCCIO and CPU System agent voltage, and increase both of them to 1.1-1.2v, try the lower settings first, then increase to 1.2 if necessary. These are voltage settings that directly affect memory operation and are often 100% necessary in order to stabilize very tight timings or higher clocks on DDR4.

I'd start with 1.36v, along with those two other changes, and run Memtest again. You can go as high as 1.38v (Actually DDR4 is generally within tolerance all the way up to 1.4v, but I would not personally take them that high just to gain higher clocks. It will significantly increase overall CPU temperatures the higher the memory clock is and the higher the memory voltage is)

Aside from that, it looks pretty good. I didn't see any mention of what your max temps were when running Prime Small FFT though. As long it was below 80°C then it's fine. If not, the overclock needs to come down a bit.

 

[Celach]

There are a few things you can do to help improve the memory performance, and then retest if you wish to.

Find the settings for VCCIO and CPU System agent voltage, and increase both of them to 1.1-1.2v, try the lower settings first, then increase to 1.2 if necessary. These are voltage settings that directly affect memory operation and are often 100% necessary in order to stabilize very tight timings or higher clocks on DDR4.

I'd start with 1.36v, along with those two other changes, and run Memtest again. You can go as high as 1.38v (Actually DDR4 is generally within tolerance all the way up to 1.4v, but I would not personally take them that high just to gain higher clocks. It will significantly increase overall CPU temperatures the higher the memory clock is and the higher the memory voltage is)

Aside from that, it looks pretty good. I didn't see any mention of what your max temps were when running Prime Small FFT though. As long it was below 80°C then it's fine. If not, the overclock needs to come down a bit.

kk, i'll start messing with those settings and retest in a bit. As for temps, during prime95 small fft they stabilize low-mid 70ºC's, and while gaming i almost never go past 60 - 65ºC, so i'm good in that regard. One thing though, when i was starting to test my overclock and reached 1.34v without touching Load Line Calibration my temps during prime95 small fft were at the 80ºC's nearing 90ºC, while after changing LLC from auto to level 4 they went down around 10ºC, which i cant really understand since i thought they were meant go up a little bit from what i researched.