Help? BSOD/Crashes 32GB RAM (4x8GB)

Sparx · 16 Nov 2020 at 10:12

Morning guys,

Hope any of you memory wizards can help me on this... My mobo is the ASUS Prime X570 Pro and RAM I'm now using is Corsair's CMW16GX4M2Z3600C18 (2 sets of this, so 32GB - 4x8GB)

I had a problem earlier this year with my PC having BSODs consistently, apps and games just crashing and closing - which got worse, it was alleviated by removing a stick of RAM so I had a single 8GB. At first I actually thought it was my motherboard going ka-put on the memory controller or something? I asked ASUS and after 2 weeks they agreed to RMA, but part way through I thought maybe it's the RAM?

So I contacted Corsair who arranged an advanced RMA quickly. This didn't resolve the issue. But not wanting to wait for ASUS horrid slow RMA service and be without a PC, I bought the same Corsair style RAM but the 'AMD Optimized' edition (this one) and it seemed to have resolved my memory issues! So I left it...

Now I've just bought another of the same AMD optimised kit, so running 32GB (4x8GB) and I noticed Discord started crashing/reloading itself (RAM dumping it?), Chrome would randomly crash some tabs with "Status_Access_Violation" errors so now and I've had a few Windows BSODs - so I'm thinking here we go... Mobo issue again?

To cut it short, I've downclocked the RAM to run 3200MHz and it appears stable the last few days over the weekend. But why should I only let it run at 3200MHz?! It's rated for 3600MHz...

Is my mobo potentially faulty, or is it I need to tweak my BIOS better? I'm not that great with manual OCs, memory timings etc so would appreciate any help or advice as to what I could do. Some quick Google-fu suggests increasing DRAM voltage, manually input timings, changing command rate to 2T and all sorts. Not sure what the optimum thing to try would be?

Appreciate any help

Thanks

chrcoluk · 16 Nov 2020 at 10:15

My first suggestion is download karhu ramtester and also hci memtest.

Do probably 2+ hour runs on both (not at same time). See if you get any errors, if no errors try it again on the rated speed. These are much better than memtest86.

MrPils · 16 Nov 2020 at 11:23

Two easy things you can try:

1) Bump memory voltage up. Its common for voltage to dip when you have 4 sticks installed. Set vdimm to 1.4v and see if that fixes it, if it does you can slowly lower it down until you find where it needs to be.

2) Also check the version numbers on the label on your Corsair kits to make sure they both use the same chips. Version 3.xx is Micron chips, 4.xx is Samsung chips, 5.xx is Hynix chips and 8.xx is Nanya. v9.xx exists as well but ICs vary, usually higher density chips so single sided 16gb sticks or 32gb sticks. I think there were some Spectek (micron oem) 8gb sticks listed under v9 as well. If you find they're different versions then swap the sticks around so the newly added ones are in the primary slots and the board reads its XMP timings from those instead. That works in a lot of cases.

uscool · 16 Nov 2020 at 12:20

As stated by MrPhils Above

I bought the same CMW16GX4M2Z3600C18 2x8GB quite some time back which are Version 3.xx Micron chips never had issues and used them on i7 6700k on Z170 Maxumus VIII Hero so pretty old now and had these running with the XMP 3600 even tightening the timings CL16

Just couple of days ago I bought another set of the same but they came 8.xx Nanya when i added them in the computer didnt Boot just blank screen so I put the 2 newer sticks into slots A1 and A2 and the older ones in B1 and B2 and it booted up and have all running 3600mhz CL16 ran few tests and had no issues so far only thing I cant set the command rate to 1T it doesnt boot but I think thats limitation how old my board and cpu

I plan on upgrading to zen3 and reading 4 sticks of single rank run slightly better and these being on offer thought i'd buy them and use on that

Sparx · 16 Nov 2020 at 14:21

The kit I replaced earlier this year in CPU-Z shows it's Micron E-die. But the new kit I received at the weekend is Nanya A-die like you @uscool

So could be worth trying the newer Nanya kit in A2/B2 and the earlier Micron kit in A1/B1 slots then I guess, and try enable DOCP @ 3600MHz?

I'll also try bumping the DRAM voltage up @MrPils too if that doesn't work - and will certainly try bench on HCI MemTest @chrcoluk tonight after work

MrPils · 16 Nov 2020 at 16:21

Sparx said:
The kit I replaced earlier this year in CPU-Z shows it's Micron E-die. But the new kit I received at the weekend is Nanya A-die like you @uscool

So could be worth trying the newer Nanya kit in A2/B2 and the earlier Micron kit in A1/B1 slots then I guess, and try enable DOCP @ 3600MHz?

I'll also try bumping the DRAM voltage up @MrPils too if that doesn't work - and will certainly try bench on HCI MemTest @chrcoluk tonight after work

You essentially want to swap the kits over Sparx so the newer sticks are in the slots that your older kit has been in. When using 4 sticks the motherboard will read auto timings from the sticks in the primary slots, it wont read timings from all four.
In the majority of cases where you're facing difficulty with otherwise "matched" kits that don't want to run together it will be a single timing that your newer kit wants set looser than your original kit sets automatically. In this case swapping the kits over in the slots so the Auto values are set by the misbehaving kit will mean your motherboard picks up that looser timing and applies it to your older kit - looser timings than stock will always work, tighter timings will not always work. A bump in memory voltage can also resolve this as well as motherboard voltage drops, hence my first recommendation being a voltage increase. In this situation though as uscool has the same kit and has confirmed that the fix is to swap over you should try swapping sticks first and increasing voltage second.

uscool · 16 Nov 2020 at 16:44

Sparx said:
The kit I replaced earlier this year in CPU-Z shows it's Micron E-die. But the new kit I received at the weekend is Nanya A-die like you @uscool

So could be worth trying the newer Nanya kit in A2/B2 and the earlier Micron kit in A1/B1 slots then I guess, and try enable DOCP @ 3600MHz?

I'll also try bumping the DRAM voltage up @MrPils too if that doesn't work - and will certainly try bench on HCI MemTest @chrcoluk tonight after work

I just put the newer ram in A1 and A2 slots and the older in B1 and B2 not A2 and B2

Sparx · 16 Nov 2020 at 16:59

uscool said:
I just put the newer ram in A1 and A2 slots and the older in B1 and B2 not A2 and B2

My ASUS board states A2/B2 should be used as the first 'primary' slots to use, then A1/B1 to use as secondary. As the RAM slots go in order of B1, B2, A1, A2 (left to right).

So I'll give that a go, as it's the only other combination I can swap too lol. Currently my earlier Micron kit is in the main A2/B2 slots, with the new Nanya kit in secondary slots A1/B1.

MrPils · 16 Nov 2020 at 17:16

Sparx said:
My ASUS board states A2/B2 should be used as the first 'primary' slots to use, then A1/B1 to use as secondary. As the RAM slots go in order of B1, B2, A1, A2 (left to right).

So I'll give that a go, as it's the only other combination I can swap too lol. Currently my earlier Micron kit is in the main A2/B2 slots, with the new Nanya kit in secondary slots A1/B1.

Yeah there's no consistency in labelling between motherboard manufacturers and slot numbering so just swap them opposite to where they are now - the important part is that you swap them as pairs so keep kit 1 in slots marked 1 and kit 2 in slots marked 2 (or vice versa) and make sure the "misbehaving" kit is in the primary pair of slots.

EsaT · 16 Nov 2020 at 18:03

A and B are memory channels.
You want identical DIMMs in same number slots of both.
Also both slots of single channel always use/share same wiring.
So who knows what memory controller takes as primary source of timings...

And usually slot number 2 is the one which should be populated first, because it's in end of the wire and has hence cleaner signaling than slot in the middle of wire.
(which suffers from signal reflections especially if end of wire is "hanging in air")
That's why two bigger DIMMs instead of four smaller is better when wanting higher clock speed/better latencies.

For four DIMMs better wiring would be rarely used "T-topology", in which wires from memory controller go to in between slots and then split to both slots with equally long wire.

MrPils · 16 Nov 2020 at 18:17

EsaT said:
A and B are memory channels.
You want identical DIMMs in same number slots of both.
Also both slots of single channel always use/share same wiring.
So who knows what memory controller takes as primary source of timings...

And usually slot number 2 is the one which should be populated first, because it's in end of the wire and has hence cleaner signaling than slot in the middle of wire.
(which suffers from signal reflections especially if end of wire is "hanging in air")
That's why two bigger DIMMs instead of four smaller is better when wanting higher clock speed/better latencies.

For four DIMMs better wiring would be rarely used "T-topology", in which wires from memory controller go to in between slots and then split to both slots with equally long wire.

The timings are always taken from the primary slots. Some manufacturers label the primary slots as 1 even though they are electrically speaking the second row of slots, others name them as per the physical layout of the board. There is no standard and both decisions can be argued for and against so its just a case of RTFM sadly.

T-Topology confuses this more because as you said the slots all have equal trace length, in which case its an arbitrary decision by the manufacturer as to which slot the bios reads as primary (though in all cases this should be clearly stated in the manual - again RTFM and don't assume). The only thing to disagree with in your comment is use of the phrase "better wiring" for T-Topology. T-Topology boards limit overclocking and tweaking with two sticks to the results achievable with 4 sticks - they make two dimm overclocking worse. Daisy chain configurations will clock two sticks better than T-Topology, yet usually clock 4 sticks the same as T-Topology. T-topology is simpler for the user when adding another 2 sticks to an existing pair, but is limited in terms of two stick tuning and clocking.

Sparx · 17 Nov 2020 at 16:58

Just as an update folks, I swapped the kits round as mentioned above - reset BIOS to defaults - enabled DOCP @ 3600MHz - and I've not had any issues since... *touch wood*

Spent a few hours playing Valhalla last night, browsing forums/Youtube the usual, and been working from my PC all day today and no issues at all. Neither Discord or Chrome crashed... so it really may have been THAT simple to just swap the kits round! Who'd have thought...

Well I'm pleased as I won't need to RMA my board now hopefully! Thanks for your help guys

MrPils · 18 Nov 2020 at 02:13

Sparx said:
Just as an update folks, I swapped the kits round as mentioned above - reset BIOS to defaults - enabled DOCP @ 3600MHz - and I've not had any issues since... *touch wood*

Spent a few hours playing Valhalla last night, browsing forums/Youtube the usual, and been working from my PC all day today and no issues at all. Neither Discord or Chrome crashed... so it really may have been THAT simple to just swap the kits round! Who'd have thought...

Well I'm pleased as I won't need to RMA my board now hopefully! Thanks for your help guys

You're welcome, glad its all sorted!

The blame for this lies squarely at the feet of corsair really, the model numbers of the kits should be different. They use this as an excuse to tell you that you should get rid of your original kit and buy a matched set of 4 off them instead. Just greedy and a thoroughly anti-consumer attitude when some simple guidance or incrementing model numbers when memory chips from different manufacturers are used in the same speed kits so users dont buy mismatched pairs of memory resolves the problem.

Jack Franklin · 3 Jan 2021 at 18:42

Put two identical sticks of dram (matching date codes) in A2 and B2 then boot to bios. Go to the bios screen where you can manually enter CAS and other latencies. You may see something like 18-22-22-22-42 in the 5 top entries. Bump the latencies up by one or two. Also bump the dram voltage up a little to say 1.45v but not above 1.5 volts. See if this helps. If the machine doesn't run properly bump the latencies up anther point.

If the machine runs OK, boots and no BSOD, put the another sticks of dram into A1 and B1 and see if the machine runs OK. If so, drop the DRAM voltage and see if the machine runs properly.

zx128k · 4 Jan 2021 at 00:54

You can't OC the RAM frequency to 3600MHz fill all slots on the motherboard and then expect it to just work (it can just work if you are lucky). With all slots full this can affect the maximum frequency possible and the stable timings of the RAM. Also two dual channel kits might not work well together. For four sticks the maximum frequency can drop under 3200MT/s.

You need to run memtest with 8 full passes and 32 passes of test 7. This way you get full coverage of all patterns in memtest for each test. It can take 8 hours, and must be run all at once concurrently. Shorter tests can miss errors.

More information about the optimal number of passes can be gleamed from the article MemTest86 Technical Information from the description of the MemTest86 config file, mt86.cfg, available only in the Pro version:

PASS1FULL

Specifies whether the first pass shall run the full or reduced test. By default, the first pass shall run a reduced test (ie. fewer iterations) in order to detect the most obvious errors as soon as possible.

MemTest86+ needs to run for at least 8 passes to be anywhere near conclusive, anything less will not give a complete analysis of the RAM.

If you are asked to run MemTest86+ by a Ten Forums member make sure you run the full 8 passes for conclusive results. If you run less than 8 passes you will be asked to run it again.

Test 7 [Moving inversions, 32 bit pattern]

This is a variation of the moving inversions algorithm that shifts the data pattern left one bit for each successive address. The starting bit position is shifted left for each pass. To use all possible data patterns 32 passes are required. This test is quite effective at detecting data sensitive errors but the execution time is long.

https://superuser.com/questions/547822/how-many-passes-are-enough-with-memtest

Jack Franklin · 4 Jan 2021 at 04:04

yes, you are correct. but the immediate problem is to get the machine running in a stable manner without with BSOD and with 4 sticks DRAM.

zx128k · 4 Jan 2021 at 08:11

Jack Franklin said:
yes, you are correct. but the immediate problem is to get the machine running in a stable manner without with BSOD and with 4 sticks DRAM.

Would be worth checking that you have your DRAM voltage set correctly and primary timings set correctly as per XMP or on the front of the RAM chips then test. Note there could be fault with your RAM. You need to rule that out. Dont mix your RAM kits up. You should make sure the RAM is not faulty by testing each kit first, then install both kits and test.

You can try setting the primary timings to be very loose. See if that helps and at some point you become more stable. Memory overclocking is a pain. Its change one timing and test. Then repeat. Once you think you have got good result then try a full memtest.

Memory Controller Limitations: While AMD’s Ryzen 3000 platform is beastly and supports extremely high memory speeds, the memory controller does still have limitations, and those limitations become noticeable when populating all four DIMM slots with either single or dual rank memory.

4 x 2 Rank safe could be 2933MHz
or
4 x 1 Rank safe could be 3600MHz

Its a try and see kind of thing. Thaiphoon Burner will be able to tell you if the RAM is SR or DR.

Part number CMW16GX4M2Z3600C18

XMP/DOCP: 18-22-22-42 @ 1.35v

So your RAM could be Micron E-die, single rank. So the safe frequency could be 3200MT/s and if luckly above that. If you fill at slots that can affect maximum frequency as well and you are hoping for the IMC to like it as well. Four slots will normally act like Dual Ranked to the IMC. Some motherboard wont support 4 x DIMMs of the same RAM. Only 1 or 2 sticks maximum.

Memory QVL for your motherboard

https://dlcdnets.asus.com/pub/ASUS/...QVL_for_3rd_Gen_AMD_Ryzen_Processors_X570.pdf

Your motherboard memory QVL shows your motherboard does not support CMW16GX4M2Z3600C18 (Ver3.31) configured as four DIMMs. This is five up from the bottem of page 4/18. 1 and 2 DIMM cofiguration are supported. So you might not be able to get four DIMMs working. CMW16GX4M2Z3600C18 also has IC's made by Spectek on page 3/18, they to are not supported as 4 DIMMs.

https://www.corsair.com/corsairmedia/sys_master/productcontent/Ryzen3000_MemoryOverclockingGuide.pdf page 11. The frequency of the RAM is reduced if all four DIMM slots are full and if the RAM is DR.

This is also the same for Ryzen 5000 series cpu's. https://dlcdnets.asus.com/pub/ASUS/...for_AMD_Ryzen_5000_Series_Processors_X570.pdf page 16/19

So this is likely a compatibility issue with your RAM in a 4 x DIMM configuration, with the PRIME X570-PRO motherboard.

DRAM Calculator for Ryzen
https://www.techpowerup.com/download/ryzen-dram-calculator/

The DRAM calculator for Ryzen can help you get started with tuning your RAM. Note here you are not looking to overclock. So stay at defaults for voltage.
I would just try to see if the primary timings can become stable. If they wont then drop the RAM and IF frequency. Only use the voltage your RAM uses by default via xmp.

Also dont go mad and over tune the memory or you will get problems down the road. Just primary and maybe secondary timing.