Can't pass POST @ 4000mhz DDR4 - Z390

I've got some new RAM (F4-4000C16D-32GTZR)

Rated for DDR4-4000 CL16-19-19-39 1.4V

The system boots happily into Windows at DDR 3800mhz at XMPII settings with VCCIO and VCCSA both at 1.1V

But no matter what I do, I can't get it to pass POST at 4000mhz. I've tried

VCCSA to 1.35v
VCCIO to 1.25v
RAM voltage to 1.45v
Loosened primary timings

Feels like I'm missing something. In fact it wont get past POST at even 3866mhz.

 

Guess it will need even more vccsa and vccio then aka you hit the wall on cpu imc and or z390 limitation in general.


OR, look at ODT timings, ondie termination voltage and set them ohm resistance voltages accordingly. Inspect what it uses at 3800 and use those, but disable any ODT training then.

Usually for samsung b-die
Wr 80
Nom 48
Park 40

On both channels. Nom and park are by some different and it's actually park 48 and nom 40.

Also park / nom value goes from 30-48, some work at 35 or 40 too..

-- and disable any odt training afterwards, or it could break and spit errors in tm5 tests


It's how I fixed 4400mhz boot on my z490, auto wouldn't boot at all as soon as I entered in those ODT vaules, bingo.

And worth mentioning, sometimes it's better to run lower strap and oc that via blck to desired speed, in my case native 4400 strap is 100:100 and that stresses imc more then 133:100 strap aka 4266mhz, so I used that and oc'ed blck to 103,2mhz and have now 4404mhz, this also made higher bandwith then native 4400.

 

might be a limitation on 9900kf
I have a 10700F with a 4133 c16-16-16-31 (oc) kit and it runs fine with vcsaa around 1.3v

 

Thanks for your help guys but I have decided to throw in the towel on 4000mhz

I ran a couple of benchmarks, my old 3600 C18 RAM is faster than the new stuff at 3800 C16

I can only think that this is because it's 4 x 8gb vs 2 x 16gb dual rank. I suspect my mobo (ASUS Prime-A) is optimized for 4 sticks as that is what was prevalent when Z390 was current.

Even if I did get the new stuff to 4000mhz, it would probably only be the same speed as the 3600mhz is now.

So I'm going to jack up the voltage on the old memory and see how far I can push it I expect the info you provided will be helpful for that @-Tj-

 

32Gb is much, much harder to IMC than a 16Gb kit. 2x16Gb slightly easier than a 4x8Gb.

32Gb kit never run as fast as 16Gb (2x8Gb).
I have 4x8Gb (4600c18 B-Die) but IMC only handle them 3800 medium/high settings or 4000 at medium settings.

Years i have used only 16Gb (2x8) becouse these rolling now ultralowlatencies.

36ns Aida64 latency test and PC is faster now, snappier, more FPS. Some games benefit "ULL" latencies and some not.

IMO 32Gb on gaming is gimmick, still, late 2022. A couple Simulators benefit 32Gb ram (MSFS) but i dont play those.

16Gb>32Gb might get only 1-2% fps increase, example 150fps to 153fps.

No way point to go 32: you lose more fps when you must drive those rams much slower bandwidth and higher latencies.

99% games, example all AAA games runs fine 16Gb ram (+virtual ram/page file on SSD) and those need only plenty of Vram becouse todays beautiful graphics + 1440/2160 resolution.

Absolutely you might get your kit to work 4000 16-18-18, but on my rig, those latencies arent acceptable. :/

 

Well this is funny. I managed to get my old 3600 C18 4x8gb stable (so far) at 3800mhz with just a bump of voltage up to 1.4V

I also got it past POST at 4000mhz by giving it 1.45V and loosening timings. It wouldn't quite boot into Windows but I'm sure that could be fixed with some tweaking.

So I know that in theory my CPU and motherboard can do 4000mhz, it must've been some sort of incompatibility with the b-die. Maybe I could've fixed that with more tweaking.

 

The issue is not the amount of ram. But the amount of populated dimms.
4 dims is much harder to run than 2 dims. Especially if they are dual rank.

 

On my system I have seemingly opposite.

4000mhz (2x16GB b-die) wont pass POST

4000mhz (4 x8GB) passes post but doesn't get into Windows.

Also, 4 x 8GB is faster at 3800mhz despite higher latencies.
3800 C16 vs 3800 C18.

 

The kit 2x16gb you posted is dual rank.
I don't know about the other.

But the reason for stability and performance difference is probably due to secondary and tertiary timings.

 

Yes, the 2x16 is dual rank. The 4x8 is single.

The 2x16gb has better primary timings all accross the board. 16-16-16-19 vs 18-22-22-42. Not sure about secondary, they were on auto so maybe....

It wasn't an insignificant difference in speed. In super pi, the 4x8gb does 1M in 7.6sec. The 2x16gb is 8sec.

 

Z370 and Z390 boards have T-Topology where 4x single rank DIMMs will work at higher freq than 2x dual ranks DIMM, just look at the memory QVL.

From the timings your 2x16GB kit use Samsung B-die while your 4x8GB kit use Micron. You can tweak the tRFC and tREFI on samsung b-die, just set tRFC to 300 and tREFI to 65676, easy 10% lower latency.

 

Yeah forgot about that, mine is daisy chain and likes DL more then SL, also it has to use A2 B2 slots to be the most efficient.


@southamptonfc

which slots did you try that 2x16GB kit, A1 B1 or A2 B2?

 

@-Tj- Yep I used the A2 B2 slots.

I just tried some Micron (Corsair) 4600 C19 single rank, 2x16gb RAM. It works much better.

It's stable (so far) at 4000 C17. I've had it up to 4400 but VCCSA was 1.35V and RAM voltage 1.45V which is a bit higher than I'd like.

At 4000mhz, it's slightly faster than my 4x8gb 3600 C18 RAM.

So I guess I need to get another 2x16gb so I can run it in dual rank I don't need 64gb RAM but it's all in the name of science. I have learned a lot about memory in the last few days!

I guess Z390 just doesn't get along too well with dual rank modules, or maybe Samsung b-die but it certainly seems to like this high density Micron.

 

OK so I got 2 more sticks of RAM.

Now running at 4000mhz, 17, 20, 20, 39, 630. That's as tight as I can go with 1.4V on the RAM.

Aida shows my memory bandwidth has gone up a lot vs 3600CL18 but games benchmarks only show a small increase in FPS.

Is it just placebo or can I notice a significant improvement in smoothness in games now? Particularly when I bumped up tREFI to 65000?

 

I guess RAM`s bandwidth is not critical for games.

 

It's a very complex equation. Resolution, GPU, type of CPU, game all have different requirements.

Then, are you looking for 60FPS or 240FPS? How about 1% lows, .1% lows.

My impression is that .1% lows have significantly improved with faster RAM but I have no evidence to back that up. Something I never knew about, tREFI feels like it gave the best improvement in feel of the games I play.

 

Couldn't get my Msi B550m mortar to boot with 4 sticks of TridentZ Royal 4000 cl15... only 2, while 2 sticks of 8gig(16) was more than enough... I paid over £400 for those puppies so went back to my x570 Unify after buying a Lian li RB-001X Radiator Bracket kit for my Lian Li mini snow case so i could fit my Unify with my Galahad 360 aio in the top of my case.

 

I also failed to utilize 4x8GB sticks at 4000MHz on my x299 rig, so I went for much tighter timings at 3733MHz.
But in my case rig booted to Windows but then - BSODs on high load.

PS I will never dive into tens of RAM settings in BIOS. I am content with current state.

 

Do not forget about cache in CPU. When you load new level in game then we can assume that cache is not involved, but after that you do not have to read RAM soon. (Simplified view but...)

 

That's normal. Every serious benchmark I can find that tests how much improvement you can expect between lower clocked and/or higher latency RAM vs high-end RAM always shows marginal differences. I settled for relatively decently priced 3600 and C18 and called it a day. I'm not gonna pay 100 bucks extra for a 3% improvement.