Hello and thank you for your suggestion 
I’m running my RAM at JEDEC 3200 so that HAS to work, the RAM is validated by the manufacturer, in this case Samsung, to work at those timings, in this case CL22 and so on.
There is no OC going on, no XMP (which is also an OC).
But you do give me the idea that I may still need to check if I have memory erros, I haven’t run memtest yet on these DIMMs.
technically if fully populated this is not true, 2933 is the max the threadripper will support…
Uhm… I only use 4 but DR would that be the case then too?
Yes
No I don’t no worries, I have 16GB DIMMs with chips on both sides, they are dual rank and I have quad channel
3200 cas 22? thats some real loose timings mate.
typically for 3200 your looking at cas 14-16.
and even without xmp or a mem profile set at 2133 the cas timings wouldn’t normally be set to 22.
can you post a screenie of your jdec from cpu-z? both the memory and spd (with an active channel selected) tabs?
CAS 22 is a JEDEC specification for 3200 that is what some ECC and normal “non-tuned” DIMMs chip as.
As I suck at explaining things I will refer you to Wikipedia, check the table further down on the right DDR4-3200AA is the standard name for it.
Did you get this sorted out? I just setup a 3960x machine and mine seems to be very stable so maybe I can help you out.
Not yet, I need to send in my CPU for the time being but that takes some time as I have to setup a second PC which can be used in the meantime, since the retailer I bought it from has no option to send me a CPU which I could use in the meantime when I have to wait for them to “check” my CPU.
How long did you have the system now for?
I finished assembling it a few days ago. So basically I have been using it this week. Replaced my 3900x / x370 board with this TRX40 MSI Creator & 3960x.
Where have you been running into issues specifically? Is it just generally or what?
update … i should have used this sooner.
ran sfc /scannow no integrity errors
first time since building this about 3 years ago.
not had a bsod all month and dism came back healthy.
seems trying to help the auth, i helped myself
…
if your still struggling. i did more digging and came up with…
set your ram to 2933 anything faster switches the mem controller over to 1 dim per channel. (wasnt sure if this was still a thing on ryzen but it is)
meaning if your fully populated and have all dim slots populated with 3200 you will get crosstalk when the ram controller tries to access 2 dims on the same channel.
2933 will also mean you should be able to tighten cas latency,
so other than loosing 266 mega transfers. no real impact.
if your lucky you might be able to get cas 20 and real lucky cas 19.
which will be a real improvement performance and system feel wise.
The issue are random bluescreens with the message “NMI_HARDWARE_FAILURE”. They can’t be reproduced and happen in intervals from a few days to months.
Could you point to a source? Maybe an article, so I could read that? I only have 1 stick per channel though.
Not much improvement honstly, but overclocking and tightening timings will only come after I can confirm that the system is stable on default settings.
Crosstalk? That is interesting too… Never even heard of that. I have 128GB and 4 Dimms (part of how i went to TR because I couldn’t make this setup work past 2800mhz on my 3900x).
However my system has been 100% stable… its honestly too fast, I just don’t like how my max temps hit 86C, even if the average is 65C
What’s your chipset temperature like?
Chipset 69C max, 66C average according to HWINFO.
but “motherboard” is 50.1 Avg & 54.3 Max
Then the “VR Mos” is 49.5 Avg and 53 Avg and
the “Chipset 2” is 58.2 Avg and 61C Max.
“External 2 and External 3” are also similar with 49.9 and 49.3 avg and 53/53.5 Max.
I also have another category that has “Motherboard” again with 51.2 Avg and 55 Max… This motherboard seems to have about 50 temp sensors…
Wow that’s nice and low, my chipset temp is going to 93C, with my new board, and then it’s throttling, since when it reaches that it suddenly goes down about 10C.
It was 80C with my old board.
this is gigabytes epyc. page 5 and 6
i first run into it in the ddr3 days with the intel i7’s you could run ddr3 1600 ram, but it would only run 3 dims if you went over 1333 as that was the native speed of the mem-controller.
seems the same rules still apply today.
i thought things might have changed with them switching to quad and octa channel configs but apparently not. 
What board do you have? I have a MSI Creator and my case is a Thermaltake Suppressor F51 I have 7 fans in here including the radiator.
Very useful information, I wondered why I hit such limitations on my AM4 platform. However I wonder if it matters if the memory is high density / Dual Rank or Single Rank? Since that makes a difference for Threadripper and AM4; it says so in both of my motherboard manuals.
How many dimms + whether or not it is dual or single rank. My 32GB sticks here are dual-rank so I can only have 1DPC vs single rank I can have 2DPC
I never had that happen, I’m even currently having one system with 4 DIMMs, Dual Rank, Dual Channel, with 1866 on an old i5. So the memory controller is fully stressed, basically to the max and the system runs without problems.
The board is an ASUS Zenith II Extreme Alpha and the case an old Lian Li PC-V700 (squezed that in there :)). Sadly only 1 140mm fan intake (which failed recently so I have to replace that now too), so I leave a side panel open on the hotter days. But even with both side panels open and basically unrestricted airflow the chipset gets hot af.
Since I had to send in my CPU as the next step to get this friggin error fixed I took the board apart to check the thermalpads on the chipset cooler.
They look fine but are probably bad ones so I’ll replace them with better ones and hopefully that helps a bit. Nobody seems to have ever done that though so I had to measure the pads myself.
If anyone wants to know the thickness of the pads then here:
Front (Where the cooler is directly on the chipset): 1,5mm
Backside (Between motherboard and backplate): 2mm