Overclock.net › Forums › AMD › AMD CPUs › [Official] AMD Ryzen DDR4 24/7 Memory Stability Thread
New Posts  All Forums:Forum Nav:

[Official] AMD Ryzen DDR4 24/7 Memory Stability Thread - Page 79

post #781 of 1143
Quote:
Originally Posted by Keith Myers View Post

I believe you should up both/either/or ProcODT and CAD Bus lines in the memory section. Try 60 ohms in ProcODT and 30 ohms in the CAD Bus driving strength resistances. I don't think you need to go that high in memory voltage just for 3200 Mhz in the first place. Somewhere around 1.38-1.40V should be sufficient even with 32 Megs. Also Ryzen Timing Checker is your friend. I would start with the Stilt's Timing charts first.
Stilt's Timing Charts

I'm trying to just straight out get 3333mhz stable now, I noticed that in many games like far cry 3 has awful big difference if I don't set affinity to use cores from one CCX (tested with RAM at 2800mhz) FPS difference is almost like 20-30FPS and that is in under 100FPS! so it's huge. Giving affinity to 4 cores (without SMT cores) gives in top of that about 5-15FPS... And total GPU usage increased from 40-70% to 70-100% with setting affinity to 4 real cores from one CCX! I wanna get max possible infinity fabric speeds with faster RAM so infinity fabric bottleneck would get smaller. GTA V is not really affected by infinity fabric because my GPU usage is almost all the time 99% but with some drops here and there probably caused by infinity fabric so probably cores can't communicate fast enough.

Here are my settings for 3333mhz
Warning: Spoiler! (Click to show)

procodt = 68.6ohm

Ram voltage = 1.4v

vddp = 0.9v

VDDSoC = 1.02v, LLC 3 (spiking to 1.05v)

vpp_mem Voltage = 2.56v

VTTDR voltage = 0.75v

RESTISTANCES

Clock Drive Strength = 40.0Ohm
Address / Command Drive Strength = 40.0Ohm
CS / ODT Drive Strength = 40.0Ohm
CKE Drive Strength = 60.0Ohm

I can run about 40-1hr prime95 BLEND test with these settings and I know if I change CKE drive strength to 40ohms from 60ohm blend test fails immediately

I think that CDS and address strength change from 32ohm to 40ohm and DRAM voltage change from 1.41 to 1.40 did not change a whole lot ( or I'm testing that change now, I bet that failure comes at about 1hr mark)

And I'm using prime95 blend first and verify that with memtest. In my use case, prime95 fails faster if RAM's are unstable.

EDIT: changing two first mem bus resitances from 32ohm to 40ohm and lowering dram from 1.41v to 1.4v allowed me to pass 1hr mark of blend, at least I'm going for the right direction. I think biggrin.gif

Do CLDO_VDDP voltage do more than just resolve memory holes? Can right CLDO voltage lower the need of vcore?

EDIT: error was detected in about 1.55h mark now, testing with memtest little bit before changing resitance(s) and trying again just to get feel of it how stable memtest is. Lol this time around memtest caught error in like 5% of coverage have to figure out what's really is really bottlenecking RAM :/
Edited by WarpenN1 - 9/23/17 at 4:53pm
post #782 of 1143
Well I am not a gamer so anything related to that is over my head. I see a couple of things that draw my attention. First is VTTDDR is disobeying the rule. It should be 1/2 Vddr. So with your Vddr set at 1.4V, VTTDDR should be at 0.7V NOT 0.75V.

The other one is that stability in higher memory clocks has been observed with setting Vddp to equal Vsoc or slightly less. That should be at ~ 1.0V with your Vsoc at 1.02V. You have it too low at 0.9V Gupsterg is the expert here.
post #783 of 1143
Do you guys think I would see any significant improvement returning my current Samsung b-die 128gb tridentz (rated at 3200mhz 15-15-15-35, but currently running at 3060mhz 14-14-14-30) for a 128gb kit that actually is rated to run at 14cas?


I'm thinking it's not really worth paying restocking fee of 10% just to try to get 128gb 14cas tridentZ when my kit already can run at 14cas. but maybe it would help me stabilize 3200mhz?
post #784 of 1143
Quote:
Originally Posted by Keith Myers View Post

Well I am not a gamer so anything related to that is over my head. I see a couple of things that draw my attention. First is VTTDDR is disobeying the rule. It should be 1/2 Vddr. So with your Vddr set at 1.4V, VTTDDR should be at 0.7V NOT 0.75V.

The other one is that stability in higher memory clocks has been observed with setting Vddp to equal Vsoc or slightly less. That should be at ~ 1.0V with your Vsoc at 1.02V. You have it too low at 0.9V Gupsterg is the expert here.

Hmmm then this is super weird but when I set my vttddr from 0.75v to 0.706v memtest fails immediately. I have DRAM voltage at 1.41 so 0.7062v (can't set it more precisely) should be about half :/

And with loosened timings 16-16-16-34-50 got me 200% coverage in memtest before fail compared 14-14-14-28-42 when it failed in couple of minutes.
Edited by WarpenN1 - 9/23/17 at 7:53pm
post #785 of 1143
Quote:
Originally Posted by mus1mus View Post

Try adding SOC Voltage.
Tune CLDO_VDDP
Add a bit of VDDP - best to match it or slightly lower than VSOC.

SoC I can play around with
CLDO_VDDP as well.

Not sure what VDDP is, MSI board may not have it.

But it seems ProcODT has a -big- influence. In an attempt to get 3200 andhgiher running, turns out 80 ohms and higher leads to 3 long beeps, automatic reboot, repeat 5 times, then it posts with RAM at default settings (Volts and CPU specific settings are still intact).

68.3 ohms though, 3200 posts right away, as does 3333. 3466 has -never- booted before, but now magically boots. It is a little problematic but I have it posting at 60 and 53 ohms (53 had problems POSTing but running GSAT now without issue, going on over an hour so far). I thought higher meant more supportive for overclocking but maybe for me its the reverse, kinda? I haven't tried any of the really low settings yet, but i've been following the rule that amd posted that higher settings may be more supportive.

So I guess dual rank single DIMM quad channel at 3333+ can be a thing, just have to fiddle with those specific stability settings. stressapptest on windows 10 bash shell hasn't thrown any errors yet (in fact it's never thrown an error yet on any setting since I get BSOD's instead, even if I run a full 4 hours of GSAT -W)

I'd love to get the full 3600 my kit is rated for, but testing each setting is very cumbersome as MSI x399 has very slow failure recovery (it could also be a threadripper thing too)
Edited by ssateneth - 9/23/17 at 8:07pm
post #786 of 1143
Single rank, B-die is supposed to run best at 53 or 60. 4 sticks often need bumping to 68. Double rank is another story entirely. I think you can count on one hand and have fingers left over of the number of people that have 3600 Mhz running stable. 3200 should be a given and 3466 is quite doable as based on the commentary and posts in all the memory threads. The BIOS' across all platforms, Ryzen or Threadripper and all manufacturers seem to put 3600 in fantasy land. Until AGESA 1.0.7 drops and some wishful thinking I believe, 3600 will remain a pipedream for most.
post #787 of 1143
Quote:
Originally Posted by happyluckbox View Post

Do you guys think I would see any significant improvement returning my current Samsung b-die 128gb tridentz (rated at 3200mhz 15-15-15-35, but currently running at 3060mhz 14-14-14-30) for a 128gb kit that actually is rated to run at 14cas?


I'm thinking it's not really worth paying restocking fee of 10% just to try to get 128gb 14cas tridentZ when my kit already can run at 14cas. but maybe it would help me stabilize 3200mhz?
No, just the opposite in fact. The lesser amount of sticks installed, the greater likelihood of getting higher memory or stock rated clocks running. The less amount of work the IMC has to do, the better the results.
post #788 of 1143
Quote:
Originally Posted by Keith Myers View Post

No, just the opposite in fact. The lesser amount of sticks installed, the greater likelihood of getting higher memory or stock rated clocks running. The less amount of work the IMC has to do, the better the results.


I dont think you understood my question. I am already running a 128gb kit that was originally rated for 15cas 3200mhz. It ended up being samsung b die and I was able to tune it to be 14cas 3060mhz.

I am now wondering if I had originally purchased a 128gb kit that was rated for 14cas 3200mhz, if I would be able to achieve 14cas 3200mhz.
post #789 of 1143
Quote:
Originally Posted by happyluckbox View Post

I dont think you understood my question. I am already running a 128gb kit that was originally rated for 15cas 3200mhz. It ended up being samsung b die and I was able to tune it to be 14cas 3060mhz.

I am now wondering if I had originally purchased a 128gb kit that was rated for 14cas 3200mhz, if I would be able to achieve 14cas 3200mhz.
Yes, I did. I'm surprised that RAM rated at 15 CAS ended up being B-die. The rule of thumb as always been CAS 14 guaranteed to be B-die or any RAM rated at 3333 or greater is B-die. The B-die ram rated at 3200 or greater usually can be tightened up to CAS14. I think that is holding true even at the higher densities.
post #790 of 1143
Quote:
Originally Posted by Keith Myers View Post

Yes, I did. I'm surprised that RAM rated at 15 CAS ended up being B-die. The rule of thumb as always been CAS 14 guaranteed to be B-die or any RAM rated at 3333 or greater is B-die. The B-die ram rated at 3200 or greater usually can be tightened up to CAS14. I think that is holding true even at the higher densities.


So should I purchase the cas14 128gb kit, or just stick with my cas15 128gb kit? I really want to try and stabilize 3200mhz, and I wonder if the higher quality ram 14cas kit will allow me to reach it. I mean I have samsung b die already (lucked out i guess) So maybe its not worth the attempt. Also.... im out of my warranty/return period...
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: AMD CPUs
Overclock.net › Forums › AMD › AMD CPUs › [Official] AMD Ryzen DDR4 24/7 Memory Stability Thread