Trying to get my RAM 100% stable - Advice needed! (3950x, Trident Neo)

[Nirurin]

Ok so, I am running:

3950x on a Crosshair Hero VII x470

Trident Neo 3600 18-22-22-42 64gb (4x16gb)

XMP is... just terrible. Boots to blue screen. No bueno.

DRAM Calculator is... also pretty terrible. The timings seem really screwy, I suspect it's not really optimised for 4 sticks like this.

I can get it into Windows and running RamTest1.1 fairly consistently, to the point where it's -almost- stable. As in, I've done this many times before on other boards and with other components, and unstable ram will throw up 20 or 30 errors fairly consistently. At the moment I'm getting 1 error for every 200% ish.

But I can't seem to figure out how I can fix it, as I'm already on some pretty loose timings (which I expected, I already knew that the chip wouldn't run this much ram at max-specs).

Currently I'm on 3600mhz, 22-22-22-48-62-630 and am on 2T command rate. Running at 1.5v dram (fine for Neo) and 1.2v soc (I thought the soc voltage was helping, as I got a 500% with zero errors earlier, but I just got a single error in another test so...maybe not).

Appreciate any help with this, as I've exhausted my (limited) knowledge of ram.

 

[Nirurin]

What are the options you have selected in Dram Calculator?
Dram voltage at least 1.35?

My DRAM is on 1.5v (seemed high to me, but it's within the recommended specs for Neo. Once I find a stable config I can try lowering it).

DRAM Calculator settings I used are:

Ryzen 2 Gen
Hynix CJR
V1 and Manual (ive tried both)
Memory Rank 2
3600
100
4 dimms
X470

Calculate Safe.

 

[Nirurin]

This is what I would try next.

I would test that ram on 2666 with standard timings of 16 18 18 39, at 1.35v.

The reason, it's possible there is a bad cell(s), and if it can't test 100% at stock 2666 you know there is fault somewhere. However if it's stable at standard speeds you can move on to getting ram stable at speed it's sold for.

I'm pretty sure I already did a 500% run at the stock 2133mhz speed (that's the default for ram, i believe?). After my current test ends though I will set it all back to defaults and do another test just to be super sure. I'll report back.

 

[Nirurin]

This is what I would try next.

I would test that ram on 2666 with standard timings of 16 18 18 39, at 1.35v.

I'm currently at 100% coverage on the stock settings (2166, 15-15-15-36-51-374-2T) without any errors. Going to leave it running for a while anyway, but I'd be interested to hear what you'd suggest my next step should be

 

[Nirurin]

I would do a memory test on it.

Once your happy with a setting, I would increase the speed slowly.

I would also investigate if there has been any motherboard compatibility issues with faster memory speeds, and look at the notes on any BIOS updates.

The important thing is small steps, and making sure the computer is stable after every memory change. Then you can objectively work out the point memory is becoming unstable.

ok so, while this didn't happen the last time I did a ram test (using RamTest1.1), it did actually throw an error this time!

So... I'm gonna try it again, maybe using memtest instead, but it seems like there might be a fault with the ram, as this was at completely stock settings?

Any recommendations for replacement ram, i guess?

 

[Nirurin]

Have you tried reseating the memory?

Also how clean were the gold contacts on the memory DIMM's, also was there any dust/dirt on the motherboard side of the memory contacts.

It was my old board (it previously had a set of ripjaw vs in there) so it's not brand new, so potentially. But all 4 slots were filled before so I wouldn't necessarily expect it to have gotten dirty. Worth a try though.

I'm also doing another test now, mostly defaults (2133, etc) but I've raised the voltage from Auto to 1.4, cos it seemed to be running at 1.1v dram according to hwinfo which seemed... low? Also running at 16-16-16 instead of 15-15-15.

Currently on 315% without errors. I'll run it to 500% just to see if any errors pop up, then I'll shut down and clean and reseat and then try for another 500%.

 

[Nirurin]

4*16gb dual rank sticks...
I'd hazard max memory speed should be ~2933c16
With memory voltage 1.45v and cldovddp 1000mv and vsoc 1.1v
I'd start at 2400c16 and work upwards

Reseated and cleaned and am doing another test at stock settings, but with 1.4v and 16-16-16.

It just popped an error at 139% while I was writing this.

 

[Nirurin]

What did you use to clean?

Isopropyl Alcohol and a clean cloth.

 

[Nirurin]

i suspect it's not the ram that's the issue, rather the load placed on the IMC.
dual rank CJR * 4 dimms places a lot of stress on the IMC so it'll probably start failing by the time the third dimm is placed and throw errors - but rather than the actual dimm causing the errors, it's the IMC


what do you mean by stock? 2400 (ie stock JEDEC settings) or stock 3600 ram bin speed?
as i posted earlier (iirc) CJR likes vdimm so i'd start off at 1.45v
and wrt to primary timings, i'd start off with 16-18-18-38-64 and go from there

Stock is 2133mhz, and auto everything for timings. Except I changed the timings to 16-16-16-16 and the voltage to 1.4v.

Don't know if jedec is different for different boards or something, but default for this board seems to always be 2133.

I'll try a test at 2400mhz, 16-18-18-38-64, 1.45v dram, 1v vdd, 1.15v soc.

 

[Nirurin]

odd. jedec spec for ddr4 max speed is 2400 - so in theory all ram that's rated above 2400 should be able to do 2400...but anyway that's besides the point... :/


good luck

jfyi to clarify, only because you listed 16-16-16-16 and i don't want a miscommunication,
16-18-18-38-64 =
tcl 16
trcd (rd and wr) 18 - ie: 18-18
trp 18
tras 38
trc 64

Yeh, I always write the rd and wr as separate numbers because.. I don't know why haha. I've entered it the way you describe there. trfc is 530 in hwinfo, command rate is 2T.

I'm not sure the IMC is likely to be the problem though, I've seen multiple examples of people running 64gb on the 3950x at 3600mhz or even higher.

 

[Nirurin]

Not doubting it ofc
What ram are they using?



Yeah the easiest (but most laborious) route would be to memtest each stick at 3600 individually to exclude a dimm issue

I just got an error at 314% using those test settings at 2400mhz.

I'll now try and do some tests using two sticks at a time and see if any errors show up then. Is this likely to be a CPU fault? Or much more likely to be RAM?

 

[Nirurin]

Test with just 2 dimms in at xmp 3600 or whatever you were trying for that was throwing errors with 4 dimms, if it's stable with 2 then you know it's the imc though it might be worth checking both pairs to make sure there's not a weak one hiding in there.

So would that then mean I would have to RMA the cpu?

 

[Nirurin]

Not doubting it ofc
What ram are they using?



Yeah the easiest (but most laborious) route would be to memtest each stick at 3600 individually to exclude a dimm issue

Just tried 2 sticks, and went with XMP 3600mhz, and it got to 500% no errors. Going to try the next two sticks now. Really hoping it throws an error, otherwise I don't really know what I should do next.

 

[Nirurin]

He is probably referring to a damaged memory controller (thats on the CPU), but test now with your other 2 DIMM's.

BTW there is a Intel CPU diagnostic you can try also, but again try those other 2 DIMM's first.

https://www.intel.co.uk/content/www/uk/en/support/articles/000005567/processors.html

Currently on 200% of the second pair running at 3600mhz, no issues.

The only other thing that is different, is that the ram is now running at 52C, while when there is 4 sticks it runs closer to 60C or so. I wouldn't have thought a minor temp difference like that should matter (I think its well within spec) but its the only other thing I can think of.

I'll reply again at 500%, or if an error pops.

Edit: Also, is that pdt likely to be helpful? It's an AMD chip, that's newer than the last update of that tool..

 

[Nirurin]

My bad, thinking you had an Intel CPU, it's late here! See if AMD have anything similar.

If these 2 other DIMM's pass, would you motherboard allow just 2 DIMM's in the other 2 slots?

It's possible there is an issue on the other 2 DIMM slots? Worth a shot of spraying those other 2 slots, but you must use electronic contact cleaner, just in case there is some dirt in them.

Yeh the second pair of ram has passed 500% and still no errors. This is at 3600mhz xmp.

I was just thinking that I would try using the other ram slots, just as a last ditch attempt to find the problem.

 

[Nirurin]

Yes try other ram slots if motherboard allows it.

Also don't think it's temp related as you had issues running at default speed, 2133 if I recall.

Hmm. So.

Two sticks in the other 'wrong' slots, and it didn't boot XMP. It cycled a few times, and then booted at the base default 2133. Not sure if that's just a motherboard thing cos it's the wrong slots though.

Running a test with them at 2133 in those 'wrong' slots, see what happens.

 

[Nirurin]

aaaand it threw an error.

But does that mean it's the CPU, or the motherboard?

 

[Nirurin]

It's one or the other. Things to try.

1) Do you have latest motherboard BIOS?
2) I would still spray out those DIMM slots where the memory is failing.

Other than this i'm lost as i'm an Intel person, need someone with AMD knowledge to know what to try next.

Do you have a spare AMD CPU you can try?

Actually I do still have my 2700x. So I could swap it back in and try it. SUCH A PAIN IN THE BUTT but I guess it's the best solution before going to the next step of replacing the cpu or the motherboard.

Ill make another attempt at cleaning those slots too.

 

[Nirurin]

OK well, I'll pick up some cleaner spray from somewhere (amazon I guess), but as far as this test has gone -

Didnt boot XMP in those slots again (but the 2700x probably can't run them anyway), and it still threw up an error after 160% of ram testing. So it would seem it's not the cpu, but is either the board or the ram?

Am going to test these same sticks again in the proper slots but with this CPU.

 

[Nirurin]

UPDATE:

Curiouser and curiouser. I just had these two sticks of ram throw an error back in the proper slots.... I'm doing another test now, but if it happens again then it's not the motherboard or the chip (i guess) but is the ram having a really... obscure issue?

Edit: I'm going to retest in both sets of slots, but with a bump in voltage as I'm wondering if the board is defaulting to a really low undervolt for some reason.

But that will have to be tomorrows job, as it's gotten very very late. I'll report back though. I really thought I'd solved it as a board problem, but now I'm not so sure.