1. If you're receiving a message that you are banned from the Current Events or Politics forums, it's not you specifically: those forums have been hidden for all users. For more info, see here.
    Dismiss Notice

CJR Tuning

Discussion in 'Memory' started by Butcher9_9, Apr 5, 2020.

  1. Butcher9_9

    Butcher9_9 Member

    Joined:
    Aug 5, 2006
    Messages:
    2,599
    Location:
    Perth , St James
    I have been spending a bit too much timing tuning my RAM on my AMD system and I thought I my share my results to hopefully speed up the process for anyone using the same Kit (results may vary).

    My Specific kit is:
    G-Skill Trident Z Neo 3600 C16 - F4-3600C16D-32GTZNC

    The stock XMP only primary times are about as good as they get but damn the secondary and tertiary times are garbage and some of the Auto voltages are very high.

    [​IMG]

    After a few weeks of testing I settled on 3733MT / 1867 FCLK (can't get 3800 no matter how much voltage I push into the thing), 1.39v Ram, 1.05v SOC and 1.05v VDDG, 1v VDDP (Ryzen Master shows different).

    [​IMG]

    Now the parts I'm not too sure about are:
    Trfc2 and Trfc4
    TwrwrSd and TwrwrDd

    I cant find too much info into them (Ratios to other timings ect), if anyone knows anything extra let me know.

    Currently Very stable
    [​IMG]

    Ironically I found a Buildzoid video on CJR when I was pretty much done with tuning, lol.

    My timings are a tiny bit tighter, but not by much.
     
    Last edited: Apr 6, 2020
    Bolter01 likes this.
  2. Bolter01

    Bolter01 Member

    Joined:
    Oct 14, 2008
    Messages:
    646
    Location:
    Country NSW
    Yep Buildzoid is the man for memory, and yes 2ndarys can be tightened a great deal and actually show noticeable improvement (benchwise)
    I ran FLK=Mem @ 1866 (CJR) and got close to my best marks, but back to B-die @ 1800 with 16-16 timings. Seems to enjoy the tighter timings as much as increased MT. Though I have never ran V over 1.35.
    Might sound ridiculous, but I also have found that the 39X can make use of a lot of ram, 16gb+ Just a few benches and games that actually improve with 4 dimms and extra Gigs, and with that I see you have 32 FTW o7
     
  3. integralfx

    integralfx Member

    Joined:
    Mar 24, 2019
    Messages:
    55
    Location:
    Sydney
    Bolter01 likes this.
  4. OP
    OP
    Butcher9_9

    Butcher9_9 Member

    Joined:
    Aug 5, 2006
    Messages:
    2,599
    Location:
    Perth , St James
    Sweet, so might be able to get down to 1 - 2 on all 6 of those timings. I know ScL version would not go below 4, dont think I tried much lower with the other 6 though as the other Reddit guide I was referencing did not really mention them much.

    Thx
     
  5. OP
    OP
    Butcher9_9

    Butcher9_9 Member

    Joined:
    Aug 5, 2006
    Messages:
    2,599
    Location:
    Perth , St James
    Spent the night tightening everything up even more. This is as good as it gets .

    Really tried to get tRDC to run at 18 and while it was windows stable it would not pass a memtest at all ( 1.1 SOC , 1.45 Ram, bumped On Die Termination and SoC LLC and everything, no luck) . Did give a large drop in latency while it was on. Oh for B Die I guess.

    Still need to do an overnight memtest.

    [​IMG]

    I hope this helps anyone with similar Ram.

    SoC is down to 1.04 now, Anything over 1.4 Mem V does not seem to do anything (Might try to go down abit).

    Trfc2/4 are higher now as they are set to Auto as Recommended above. Cant see any difference in performance with them being way higher so I guess that advise is correct or at the very least they are insignificant timing.
     
  6. integralfx

    integralfx Member

    Joined:
    Mar 24, 2019
    Messages:
    55
    Location:
    Sydney
    On CJR, CL and RFC scales with voltage and possibly tRCD (tho only a bit), which explains why you didn't see much difference with more RAM voltage. You'd probably need 1.5v for CL14 which isn't recommended for CJR and to run CL15, you'd need to disable GDM which is really difficult to stabilise.

    Your tRFC is already pretty low for CJR. Typical is 260-280ns (485-523 at 3733MHz) and you're at 254ns.
     
  7. OP
    OP
    Butcher9_9

    Butcher9_9 Member

    Joined:
    Aug 5, 2006
    Messages:
    2,599
    Location:
    Perth , St James
    Tried no GDM before I did much tightening and it was not happening let alone now. Its on a knives edge at the moment:

    tRFC , I know it will not post at 464 and gives lots of errors at 470 (was going in steps on 4 to save time), have not tried between 474 and 470 but its so little reward for the effort. Every other timing is 1 step away from Errors or not posting at least with current voltage and resistance.

    Tried 1.38VDD and 1.04 SOC but that gave errors aw well so Voltage will ahve to stay put.
     
  8. OP
    OP
    Butcher9_9

    Butcher9_9 Member

    Joined:
    Aug 5, 2006
    Messages:
    2,599
    Location:
    Perth , St James
    Update on this one, I just got a 5900X to replace my 3900x.

    At first I thought I had a lemon as it was horribly unstable at 1900 Fclock ( 1.45V mem, 1.125 SOC ect ect) , but I remembered something about some chips not liking round numbers and turns out it works just fine at 1966 or 2033. I was pretty surprised that my CJR memory went as high as 4066 (tried 4133 but it would not post, might have been the F clock though).

    I'm still at very loose timings and am just stability testing CL at the moment

    4066 = CL18 lowest (10 hour memtest stable)
    3933 = CL16 lowest ( 5 hours into memtest)

    I'm thinking 3933 with CL16 would be better but let me know if you think otherwise.
     
  9. Elmf

    Elmf Member

    Joined:
    Jan 6, 2007
    Messages:
    1,498
    Location:
    Melbourne
    My vote is 3933 CL16
     
  10. dasa2

    dasa2 Member

    Joined:
    Feb 5, 2003
    Messages:
    978
    Location:
    Northern VIC
    How about 4066 c17?
    Also try with GDM disabled on the 5900X as GDM off 2T is faster than GDM on 1T but some MB BIOS require fiddling with CAD setups to get it stable so just see how you go.

    Try these Rtt on the 5900X with ~36 ProcODT
    The DrvStr (CAD) values may help or they could reduce stability this varies with MB\sticks\PCB as to what values are best 40-20-24-24 may also be worth a go.
    [​IMG]

    Have you checked for WHEA errors when running a high IF? give OCCT a run.

    Edit:
    Here is a handy guide if you haven't seen it yet
    https://github.com/integralfx/MemTe...DR4 OC Guide.md#finding-the-maximum-frequency
    There is some great info here enough to get lost in.
    https://www.overclock.net/threads/o...24-7-memory-stability-thread.1628751/page-450

    For example...
     
    Last edited: Jun 5, 2021
  11. OP
    OP
    Butcher9_9

    Butcher9_9 Member

    Joined:
    Aug 5, 2006
    Messages:
    2,599
    Location:
    Perth , St James
    I'm still fiddling with settings, Ryzen Dram Calc sent me on a wild goose chase for a week. I had set pretty much everything to auto other than voltages and tuned the first 6 of so timing and everything was going well. Then I set all of the Resistances ect and foolishly set a few other timings and stability went to shit. I pegged it to the timings but was the other settings.

    I'll give it a good and see what happens. I would very much like to get over 4000 just because its a physiological barrier so to speak.
     
  12. Deeb0

    Deeb0 Member

    Joined:
    Mar 9, 2010
    Messages:
    206
    Location:
    3207
    This is great guys, some good info here!

    Have a similar kit to you butcher but have not had time to tinker yet. You're still on the board in your sig?
     
  13. OP
    OP
    Butcher9_9

    Butcher9_9 Member

    Joined:
    Aug 5, 2006
    Messages:
    2,599
    Location:
    Perth , St James
    Yeah same board.

    Ram overclocking is a pain in the ass , I think I'm only doing it now to keep me busy while i wait for a GPU.
     
  14. popmenow

    popmenow Member

    Joined:
    Apr 2, 2019
    Messages:
    1
    I run the exact same ram kits F4-3600C16D-32GTZNC but I am using 2 sets of them on a 5900x. This is the setting I use for them. CJR does not seem to like the tRP be lowered...
    upload_2021-7-9_18-13-28.png
     
  15. OP
    OP
    Butcher9_9

    Butcher9_9 Member

    Joined:
    Aug 5, 2006
    Messages:
    2,599
    Location:
    Perth , St James
    No, that it does not.

    I'm having a lot of troubles with my secondary timing at 3933, I get it stable , change a setting and cant get it stable then go back to the previous stable settings and its no longer stable . Then start from the beginning again. Its always a pain when it takes 20+ hours of testing for an error pop up. Never sure if its 100% stable as one could happen in hour 21.
     
    Last edited: Jul 15, 2021
  16. dasa2

    dasa2 Member

    Joined:
    Feb 5, 2003
    Messages:
    978
    Location:
    Northern VIC
  17. Audionut

    Audionut Member

    Joined:
    Mar 7, 2006
    Messages:
    1,144
    Location:
    4610
    This is what I got on an Intel board before I gave up.

    [​IMG] [​IMG]


    Some notes:
    I found the error reasons in @dsa2 post here to be accurate: https://forums.overclockers.com.au/posts/18952235/
    Appears to originate here: https://www.overclock.net/threads/memory-testing-with-testmem5-tm5-with-custom-configs.1751608/
    Code:
    KedarWolf thinks the default MT.cfg and 1usmus_v3.cfg tests might match these error numbers but the other configurations might not, so you might have to do some homework
    Here is his post
    It is said that CJR doesn't do anything with more than ~1.4v, that is my experience also.
    Mine didn't like higher speeds. I could boot at ~4100mhz but not pass tests, even with very lax timings, and trying 1.5v.
    tRCD/tRP like to be lax.
    I could probably tighten the secondary/tertiaries further, but I got the shits with rebooting, and playing with these often required a CMOS reset which only aggravated me further.

    VCCSA @1.2v - VCCIO @ 1.13v - DRAM @ 1.4v.
     
    Last edited: Jul 23, 2021
  18. dasa2

    dasa2 Member

    Joined:
    Feb 5, 2003
    Messages:
    978
    Location:
    Northern VIC
    That is where some of the biggest gains are to be had and if you hook up the case reset button to the BIOS reset and save your profiles in BIOS so they are easy to reload then it is a quick process if the system refuses to post just hit the reset button and then the power button reload the BIOS and away you go.

    There is a few little rules which I have trouble remembering.
    tRAS = tRCD(RD) + tRTP
    tRC = tRP + tRAS. Increase if unstable.
    tWR = tRTP x2
    tFAW = tRRDS x4 or x tRRDL
    tWRRD with 4 sticks or dual rank should be the average of tRCD \ 4 so tRCD 16 should be tWRRD 4 and tRCDRD 16 with tRCDWR 8 should be tWRRD 3 but this rule changes to something else with two single rank sticks.

    Most of those rules are here
    https://github.com/integralfx/MemTestHelper/blob/master/DDR4 OC Guide.md#finding-the-maximum-frequency

    And some rules are made to be broken or are just more actual guidelines but if not followed can often lead to lower performance or loss of stability.
     
    Last edited: Jul 24, 2021
    spit051261 and OCMunkee like this.
  19. OCMunkee

    OCMunkee Member

    Joined:
    Jun 28, 2004
    Messages:
    1,698
    Location:
    Melbourne, VIC
    Hey Butcher9_9 - how did you go with your CJR? I have another thread about tuning my TridentZ Neo (CJR) here: https://forums.overclockers.com.au/...jr-memory-tuning.1305166/page-3#post-18957145

    I've managed to get mine running at these timings, I get WHEA errors but related to non-critical peripherals (I think onboard LAN/Audio which aren't in use), passing TM5 with 1usmus v3 profile.

    Primary Timings Secondary Timings Tertiary Timings Voltages & Other Other
    MCLK 1967.33 tRRDS 4 tRDRDSCL 4 VSOC (SVI2) 1.1250V ProcODT 36.9 Ohm
    FCLK 1967.33 tRRDL 6 tWRWRSCL 4 CLDO VDDP 0.9297V RttNom Disabled
    tCL 16 tFAW 16 tCWL 14 VDDG CCD 0.9887V RttWr RZQ/1
    tRCD 19 tWTRS 4 tRDWR 8 VDDG IOD 0.9887V RttPark RZQ/6
    tRP 21 tWTRL 10 tWRRD 4 VDIMM 1.440V ClkDrvStr 24.0 Ohm
    tRAS 36 tWR 12 tRDRDSC/SD/DD 1/7/7 AddrCmdSetup 0 AddrCmdDrvStr 20.0 Ohm
    tRC 50 tRTP 8 tWRWRSC/SD/DD 1/7/7 CsOdtSetup 0 CsOdtDrvStr 24.0 Ohm
    tRFC 485 tCKE 1 . CkeSetup 0 CkeDrvStr 24.0 Ohm
    GDM Enabled PDM Disabled ...
     
  20. OP
    OP
    Butcher9_9

    Butcher9_9 Member

    Joined:
    Aug 5, 2006
    Messages:
    2,599
    Location:
    Perth , St James
    I gave up in the end. I got as far as 16, 16* , 19 ,19, 35, 54 for primaries and 500 for RFC , 4 and 6 for RRDS/L. I tried 16 for tFAW but it was legit not stable. 20 was super close too stable but I would get an error after about 20 hours of memtest, same goes for just about any other timing I tried, seemed so random. I would get 20+ hours stable (3Hours -TM5, 1 Hour -OCCT and 20 hours memtest) and it would be stable, then I would change one setting and 1 error in 20 hours, I would change the setting back to default and I would still get 1 error in 20 hours. So frustrating, did about 2 months of stress testing 18 hours per day.

    That being said now that you mention WHEA errors that are non memory based , it could be that I was just getting errors for unrelated devices which is why it was so inconsistent, how do you tell its not memory related?

    My Mem Voltage is 1.43, SOC is 1.1 or 1.12, other voltages are about hte same as yours (Higher did not help) . Tried ProOCT from 28 - 80 and I think 30-36 was the best but its hard to say exactly which when it test so long for errors. Drive strenghts are inline with yours.
     

Share This Page

Advertisement: