New RAM not detected

bonanza · 2019-01-30 22:33:22

loqs wrote:

CMK16GX4M2B3000C15 * 4 appear to me to 4 rank 1 modules currently running at 2133MHz@ 1.2V.
Although it is specified to 3000MHz @ 1.35V.
CPU-Z for Windows should be able to provide more information on each memory module.

I have checked CPU-Z, and it also reports 32 GB, please see: https://i.imgur.com/eCdULYyg.png

I also stumbled upon this thread: https://www.reddit.com/r/buildapc/comme … only_16gb/
where someone with same chipset X370 seems to have the same issue.

This thread lead me to check the supported RAM table of the board: https://www.asrock.com/mb/AMD/Fatal1ty% … asp#Memory
It doesn't list my particular RAM but a similar one when searching for the substring "CMK16GX4M2"

Does this potentially mean that there is no support from the main board for some 4x8GB corsair stick configurations? But why are they then recognized?

loqs · 2019-01-30 23:40:16

bonanza wrote:

This thread lead me to check the supported RAM table of the board: https://www.asrock.com/mb/AMD/Fatal1ty% … asp#Memory
It doesn't list my particular RAM but a similar one when searching for the substring "CMK16GX4M2"
Does this potentially mean that there is no support from the main board for some 4x8GB corsair stick configurations? But why are they then recognized?

Not being in the list means the motherboard manufacturer is not listing it as known to work in that motherboard.
Did dropping the memory frequency to 1866 have any effect?

bonanza · 2019-01-31 06:55:54

loqs wrote:

bonanza wrote:
This thread lead me to check the supported RAM table of the board: https://www.asrock.com/mb/AMD/Fatal1ty% … asp#Memory
It doesn't list my particular RAM but a similar one when searching for the substring "CMK16GX4M2"
Does this potentially mean that there is no support from the main board for some 4x8GB corsair stick configurations? But why are they then recognized?
Not being in the list means the motherboard manufacturer is not listing it as known to work in that motherboard.
Did dropping the memory frequency to 1866 have any effect?

I have lowered the freq. to 1866, but I am not sure whether the BIOS accepted the configuration. After I saved and restarted, the main board beeped a couple of times and then booted again.
But I could not find any change, neither in CPU Z nor in dmidecode for the configured speeds.

# dmidecode -t memory  | grep Speed                                                                                                                                                                                                                             :(
        Speed: 2134 MT/s
        Configured Memory Speed: 1067 MT/s
        Speed: 2134 MT/s
        Configured Memory Speed: 1067 MT/s
        Speed: 2134 MT/s
        Configured Memory Speed: 1067 MT/s
        Speed: 2134 MT/s
        Configured Memory Speed: 1067 MT/s

seth · 2019-01-31 08:45:37

The beeps are actually encoded status messages, 1866 is technically underclocked, so maybe it dislikes that.
Google says the board can handle 4 16GB DIMMs…

Did you btw. try memtest86?

mich41 · 2019-01-31 13:09:51

bonanza wrote:

I also stumbled upon this thread: https://www.reddit.com/r/buildapc/comme … only_16gb/

Looks like the same problem but chipset should have nothing to do with it because RAM is connected directly to the CPU nowadays.
It's entirely a matter between the CPU, RAM modules and motherboard PCB in between. And BIOS which has to bring it all up.

Interestingly, the last response has a link to Gigabyte's memory QVL and lo and behold, your RAM is there and reportedly only supported up to two slots.

bonanza wrote:

Does this potentially mean that there is no support from the main board for some 4x8GB corsair stick configurations?

It may be possible, subject to things like internal geometry of the DRAM chips or electric characteristics of the module which may introduce too much signal distortion if too many modules are plugged into one channel.
I thought we are past that kind of problems in 2019, but the QVL very clearly says otherwise.

bonanza wrote:

But why are they then recognized?

Because all the data you see in BIOS setup, DMI, CPU-Z and so on are simply copied from a small EEPROM chip on the DRAM module. It's no rocket science to read and decode that EEPROM, problems begin when the BIOS has to configure the DRAM controller of the CPU to access all that RAM.

BTW, you can check what happens if you install only two modules (on one channel and on different channels) or three modules. Maybe it will give a clue to what's going on, but from the way things look now, I think you are going to be returning that memory anyway.

Last edited by mich41 (2019-01-31 13:15:24)

bonanza · 2019-02-02 08:24:37

I have run memtest, and it also reports 16GB only.

I have contacted AsRock suport and they responded that I shall try to reset BIOS via CMOS jumpers. That's what I did, but it didn't help.
Further the support mentioned that only 16GB are recognized may be because the BIOS recognizes them as single channel only (it should be dual channel).
I then googled a bit and finally concluded to bring the RAM voltage to the mentioned 1.35V (as per vendor information). Interestingly, this happens automatically when I turn on "XMP" in BIOS.
What is totally weird to me is, if I enable XMP (which overclocks the RAM to 3000MHz and turns on 1.35V), also the BIOS beeps 3 times and then 1 time (this also happened when I try to save the RAM clock before, as per recommendation above). I then found on AsRock website https://www.asrock.com/support/faq.asp?id=2 that this means:

1 short Memory refresh timer error.
3 short Main memory read / write test error.

I then checked CPU-Z and dmidecode, and both seem to suggest that the overclocking and voltage adaptations are not applied (still at 1.2V).

Is this now hinting at a main board or a RAM error?

I also made a couple of test with the RAM arrangements:

Case 0 (as before upgrade):
A1: old1
A2: old2
B1: --
B2: --
Result: 16GB single

Case 1:
A1: old1
A2: --
B1: old2
B2: ---
Result: 16GB Dual

Case 2:
A1: old1
A2: new1
B1: old2
B2: --
Result: 24GB Dual

Case 3:
A1: old1
A2: --
B1: old2
B2: new2
Result: 24GB Dual

Case 4:
A1: old1
A2: old2
B1: new1
B2: new2
Result: 32GB Single

Case 4:
A1: old1
A2: new1
B1: old2
B2: new2
Result: 32GB Single

The 24GBs are also recognized by free and htop, while the 32GBs only show up as 16GB.

Last edited by bonanza (2019-02-02 15:46:56)

seth · 2019-02-02 08:56:25

The product description for those DIMMs stress that they're meant for overclocking and require 1.35V - apparently many (the gigabyte link supports your findings) boards cannot provide this for more than 2 DIMMs
You can try to fiddle w/ timings, voltage and clockrate (to the degree the board provides this independently) on whether you find a setup which board and RAM can handle at the same time, but I wouldn't hold my breath.

mich41 · 2019-02-02 15:25:14

Wait, wait, let's clear that "channels" thing up.
Physically, the DRAM controller in the CPU has two independent channels, usually wired to the slots like one of the diagrams below:

channel1---slot1-----------slot3
channel2-----------slot2-----------slot4

channel1---slot1---slot2
channel2-------------------slot3---slot4

You put all RAM on one channel and you have single channel operation. You put them on different channels and you have dual channel operation. The manual should show how the slots are connected to channels.
Then RAMs have those SPD EEPROMs I mentioned before, which connect through slots to the motherboard. The BIOS always knows how many modules are installed, on which channels, in which slots and what their capabilities are.
And you already saw that this information is reported correctly by BIOS even with all four modules installed, so everything is "recognized" just fine. However, the BIOS can still misconfigure stuff, disable one channel or even individual modules and generally do any number of stupid things.

bonanza wrote:

BIOS beeps 3 times and then 1 time (this also happened when I try to save the RAM clock before, as per recommendation above).

BIOSes can do it if some memory parameters change and others (set automatically by BIOS) need to be updated because memory fails to work. The BIOS will reboot and retune the hidden parameters.
If it reboots and works and reports 1.35V in BIOS menu, XMP is probably applied correctly.
If it beeps on every boot, that may be a problem.

bonanza wrote:

I then checked CPU-Z and dmidecode, and both seem to suggest that the overclocking and voltage adaptations are not applied (still at 1.2V).

DMI doesn't know about overclocking or even XMP. CPU-Z seems to read data directly from SPD and DRAM controller configuration registers, so it can report things like XMP profiles and timings configured by BIOS, but I still don't see it reporting current voltage anywhere on your screenshots. I don't think it's even possible to determine that voltage without arcane, board-specific knowledge.
It may be connected to some voltage sensor if your board has those, but you need to guess which sensor it is out of many.
A Windows overclocking utility provided by motherboard vendor may know such things.

bonanza wrote:

I also made a couple of test with the RAM arrangements:

Okay, you need to tell us where you got those slot numbers (A1/A2/...), single/dual indication and total capacity numbers from. Was it BIOS setup, DMI, CPU-Z, task manager, htop, ...?
Hard to know what they mean otherwise. As you see, those sources oftentimes don't even agree with each other.

bonanza wrote:

Case 0 (as before upgrade):
A1: old1
A2: old2
B1: --
B2: --
Result: 16GB Dual
Case 1:
A1: old1
A2: --
B1: old2
B2: ---
Result: 16GB Dual

Doubt it. IMO either A1,A2 or A1,B1 ought to be on the same channel and therefore one of those options should show single-channel.
Otherwise, both A2,B1 are the same channel and A1,B2 the other one, a numbering scheme that makes no sense at all.

bonanza wrote:

The 24GBs are also recognized by free and htop.

Nice, better 24G than nothing

seth wrote:

The product description for those DIMMs stress that they're meant for overclocking and require 1.35V

It is my understanding that the (lower) frequency and timings indicated in the standard SPD profile are supported at the standard voltage of 1.2V and the additional XMP profiles specify operation at higher voltages and allow reaching the full advertised speed.

seth wrote:

apparently many (the gigabyte link supports your findings) boards cannot provide 1.35V for more than 2 DIMMs

Unlikely, people overclock the hell out of their machines and RAM power supplies running out of juice just doesn't seem to be a problem.

Last edited by mich41 (2019-02-02 15:52:05)

bonanza · 2019-02-02 15:34:22

@mich41 thanks for your detailed reply.
I am sorry for always forgetting some important stuff. I used the same naming as in the manual, please see https://i.imgur.com/h65iLHG.png
I just assumed that slots from left to right are A1,A2,B1,B2.

I was also very confused with the dual channel mode for some configurations, but I am quite sure that this is what the BIOS reported. I will double check it.
All mentioned dual/single channel modes are those reported by the BIOS. I didn't check them in the OS because I dont't know how/where.

bonanza · 2019-02-02 15:46:34

mich41 wrote:

Doubt it. IMO either A1,A2 or A1,B1 ought to be on the same channel and therefore one of those options should show single-channel.
Otherwise, both A2,B1 are the same channel and A1,B2 the other one, a numbering scheme that makes no sense at all.

You are right, one of the entries was wrong. I have edited the post to reflect that.
Interestingly, the 24GB config is indeed reported as dual.

Last edited by bonanza (2019-02-02 15:53:42)

mich41 · 2019-02-02 17:15:28

Okay, so that makes sense now. A/B are the two channels, 1/2 are slots in each channel.

Your 24GB configuration is dual channel because both channels are populated. Some further performance improvements are possible if the channels have equal amounts of RAM in them and that's why everyone will tell you to match pairs and so on, but technically even 8+16 is still "dual channel".

The bad news is, I don't know why it doesn't work. It seems you did the right thing with old,new,old,new. Each channel gets one new and one old in the same order - it should work. And simply removing one new module makes it work indeed. WTF.

It looks like each channel individually can work with two sticks, but somehow not both at the same time because then the BIOS disables one channel. Really, WTF.

I guess this is what that customer rep said about "recognizing dual channel as single channel" but I still can't imagine why such a thing would happen. Can you ask them what conditions cause such issues? Probably their script has no answer better than "try other RAM", but what the heck, it cost nothing to ask

And I suppose you can't return the memory because old is really old and only new was bought recently?

Perhaps it is some issue with signal integrity, too many things talking to each other and disturbing others, even across channels. After all, the channels are wired close to each other on the board, some crosstalk is possible.
At this point I would be doing two things:
new,old,new,old or old,new,new,old or new,old,old,new. But I doubt it will work if both old,new,old,new and old,old,new,new failed.
Switch memory configuration to manual and set
-a few different frequencies, from the lowest possible to whatever is the non-XMP default.
-1.2V or 1.35V set manually
-1.35V and anything between the non-XMP default and the XMP default
-leave timings on auto, they have no influence on signal integrity
I have a vague recollection that I sometimes had some DDR3 quad sets which would only work at certain speed/voltage combinations.
I think one even was so mad it worked at 1600 but totally failed at 1333 which was the default on that mobo.

Some motherboards have options like "advanced DRAM drive configuration", you may try to monkey with that stuff but I've never had any success doing so, probably waste of time.

bonanza · 2019-02-02 17:20:34

mich41 wrote:

At this point I would be doing two things:
new,old,new,old or old,new,new,old or new,old,old,new. But I doubt it will work if both old,new,old,new and old,old,new,new failed.
Switch memory configuration to manual and set
-a few different frequencies, from the lowest possible to whatever is the non-XMP default.
-1.2V or 1.35V set manually
-1.35V and anything between the non-XMP default and the XMP default
-leave timings on auto, they have no influence on signal integrity
I have a vague recollection that I sometimes had some DDR3 quad sets which would only work at certain speed/voltage combinations.
I think one even was so mad it worked at 1600 but totally failed at 1333 which was the default on that mobo.
Some motherboards have options like "advanced DRAM drive configuration", you may try to monkey with that stuff but I've never had any success doing so, probably waste of time.

Thanks for your reply! I will give it a shot.
I just want make sure that (this time ;-)) I am precise: I bought the same RAM as before (its exactly the same part number) because I already was expecting weird stuff (but not that level of weirdness)

Let me maybe ask you one thing straight: Do you think its a RAM or a main board issue? (Given that in the reddit thread, the guy was also using a x370 board)
I am asking because IMHO, I have two options: Return the new RAM (1 week remaining to do it without any reasons) or return main board (2 year warranty is expiring in june)

Last edited by bonanza (2019-02-02 17:22:43)

loqs · 2019-02-02 17:23:53

Three modules might be a tolerable load at 1.2V while 4 modules the firmware decides to shutdown the second channel although I would really expect it to just not pass POST in that case.
However what does 1.35V with the rest of the settings as per the SPD for 2133MHz produce?
Edit:
I do not think it is down to just the memory or just the mainboard I think the issue is that particular combination is not working.
If you wanted to be absolutely follow the recommended settings you would need to find four single rank modules or two dual rank that can do 2400Mhz or 2666MHz @ 1.2V.

Last edited by loqs (2019-02-02 17:30:34)

bonanza · 2019-02-02 17:29:59

loqs wrote:

Three modules might be a tolerable load at 1.2V while 4 modules the firmware decides to shutdown the second channel although I would really expect it to just not pass POST in that case.
However what does 1.35V with the rest of the settings as per the SPD for 2133MH produce?

Thanks for your reply!

I just tried XMP with 3000MHz and 1.35V for two sticks only, it already starts beeping (3 beeps, pause, 1 beep).

dmidecode still shows 1.2V and this:

% sudo dmidecode -t memory | grep -i speed
Speed: 2134 MT/s
Configured Memory Speed: 1067 MT/s
Speed: Unknown
Configured Memory Speed: Unknown
Speed: 2134 MT/s
Configured Memory Speed: 1067 MT/s
Speed: Unknown
Configured Memory Speed: Unknown

So I guess with the beeping, the BIOS returned the settings to defaults.

Last edited by bonanza (2019-02-02 17:31:58)

loqs · 2019-02-02 17:40:01

3 short beeps memory error (firmware probably reset timings here) 1 long beep passed POST.
Edit:
Decrease the frequency from 3000Mhz but keep the voltage higher than 1.2V is looking to be the last possibility with those components.
Unless you want a faster configuration with two modules I would work with all four modules installed.

Last edited by loqs (2019-02-02 17:43:15)

mich41 · 2019-02-02 18:31:59

bonanza wrote:

I bought the same RAM as before (its exactly the same part number) because I already was expecting weird stuff (but not that level of weirdness)

Yes, buying the same part number is a good idea, but I don't know how much consistency those vendors maintain between different lots of nominally the same part. Do they use the same DRAM chips or perhaps change vendors? Is the internal geometry (rows x columns) in those chips the same?
Some QVLs list different versions of the same part number, what are the differences?
Are their electric properties the same? Things that normally shouldn't matter as long as they are within certain limits, but can matter when OCing? I would treat them more like two matched pairs than a matched quad.

bonanza wrote:

Let me maybe ask you one thing straight: Do you think its a RAM or a main board issue? (Given that in the reddit thread, the guy was also using a x370 board)

I can't answer that because I don't even know what's wrong.
I doubt it's a problem with any individual RAM because it would be just one of them always breaking things until you remove it.
I doubt it's a failure of any part of the mobo or CPU because most likely it would be localized to one slot.
And chipset has nothing to do with it.
Ultimately, they'll say it's your fault because you don't use vendor-approved memory

It's some stupid problem with a bunch of parts which have never been tested to work with each other. And if it's a signal integrity issue, you really aren't going to find out what exactly is wrong without some $$$$$ test equipment.

And it could be something else. Maybe the BIOS really is configured to reject some quad configs on the basis of power draw, though I doubt it. OCed setups probably draw even more power and failing RAM PSUs really just don't seem to be a problem that people have.

Do you have manual DRAM drive control?

bonanza wrote:

So I guess with the beeping, the BIOS returned the settings to defaults.

If the BIOS beeps and reboots then yes, it's probably going to change some settings.
Power off, power on until it boots without beeping. Then go to BIOS menu and see what it did. This is the settings it will be using from now until you change something or it beeps again.
I wouldn't trust DMI too much. It uses data from BIOS anyway, it can't know anything the BIOS doesn't know but it can know less.

edit: removed nonsense.

Last edited by mich41 (2019-02-02 19:09:43)

Arch Linux

#26 2019-01-30 22:33:22

Re: New RAM not detected

#27 2019-01-30 23:40:16

Re: New RAM not detected

#28 2019-01-31 06:55:54

Re: New RAM not detected

#29 2019-01-31 08:45:37

Re: New RAM not detected

#30 2019-01-31 13:09:51

Re: New RAM not detected

#31 2019-02-02 08:24:37

Re: New RAM not detected

#32 2019-02-02 08:56:25

Re: New RAM not detected

#33 2019-02-02 15:25:14

Re: New RAM not detected

#34 2019-02-02 15:34:22

Re: New RAM not detected

#35 2019-02-02 15:46:34

Re: New RAM not detected

#36 2019-02-02 17:15:28

Re: New RAM not detected

#37 2019-02-02 17:20:34

Re: New RAM not detected

#38 2019-02-02 17:23:53

Re: New RAM not detected

#39 2019-02-02 17:29:59

Re: New RAM not detected

#40 2019-02-02 17:40:01

Re: New RAM not detected

#41 2019-02-02 18:31:59

Re: New RAM not detected

Board footer