Viper Gaming 4400 C19

Discussion in 'Memory' started by M1ck1, Aug 8, 2021.

  1. PULS8R

    PULS8R Member

    Joined:
    Apr 1, 2003
    Messages:
    665
    Location:
    3029
    Wow nice work, I can't get it to run at 1T GDM off, any tips?

    Awesome, thanks to your settings I'm giving this a go. Nice rig btw, quality components.

    Umm this is what has be stumped, my ;latency isn't all that good... considering the timings. I have no idea what to tweak, but it's got to be a bios setting somewhere...

    My best latency is 55.7ns in diagnostic mode, usually up round 60 booting normally.

    upload_2021-8-15_14-45-36.png upload_2021-8-15_14-48-14.png

    I might try a setting in my bios to do with latency now I think about it.
     
  2. micsway123

    micsway123 Member

    Joined:
    Aug 7, 2020
    Messages:
    752
    Location:
    Melbourne 3015
    yea that’s a bit high? I’m getting 53.5ns with 3800cl15 tight timings.
     
  3. dasa2

    dasa2 Member

    Joined:
    Feb 5, 2003
    Messages:
    998
    Location:
    Northern VIC
    Didn't have any trouble running 1T GDM disabled with my 4x8GB 3200c14 sticks or 2x8GB of the 4400c19 but couldn't run 4 sticks of the 4400c19 at anything over 2133 with 1T GDM disabled unless I set AddrCmdSetup 56.
    It does come with a performance hit but for me it was a fair bit less than 2T or 1T GDM enabled.

    The A2 PCB used by the Patriot Viper 4400c19 and 1T are both very dependent on good MB traces to the CPU.
     
    Last edited: Aug 15, 2021
    PULS8R and micsway123 like this.
  4. glnn_23

    glnn_23 Member

    Joined:
    Jan 19, 2012
    Messages:
    555
    The combo I have runs IF 2000 with the complementary whea errors.

    Have you run a memory stress test with IF 2000 like TM5 or HCI memtest and also looked for errors in HWinfo.

    Unless you're running tight timings with it then 3800c14 or 3800c15 will probably be better.
     
    Last edited: Aug 15, 2021
  5. dasa2

    dasa2 Member

    Joined:
    Feb 5, 2003
    Messages:
    998
    Location:
    Northern VIC
    Interesting to hear with some claiming that MB doesn't get the 3800+ WHEA errors but I guess it often comes down to the CPU.
     
  6. nCrypt

    nCrypt Member

    Joined:
    Feb 12, 2010
    Messages:
    2,556
    Location:
    Melbourne
    i leave my PC on 24/7, up time currently 3 weeks, game pretty much all day every day, since cant work due to lockdown. never an issue.

    so no idea if have whea errors, if i do, doesnt affect me at all. :)
     
    M1ck1 likes this.
  7. dasa2

    dasa2 Member

    Joined:
    Feb 5, 2003
    Messages:
    998
    Location:
    Northern VIC
    It can lower minimum FPS especially if you don't stop it logging all the WHEA errors.
     
  8. nCrypt

    nCrypt Member

    Joined:
    Feb 12, 2010
    Messages:
    2,556
    Location:
    Melbourne
    Good to know.

    currently play Dead by Daylight and Overwatch on this PC, thats it, haha. since im in lockdown, will do some investigating tomorrow. :)
     
  9. micsway123

    micsway123 Member

    Joined:
    Aug 7, 2020
    Messages:
    752
    Location:
    Melbourne 3015
    dasa2 any clue what’s holding back the latency here? PULS8R should be getting closer to 52 I would have thought at that speed.
     
    PULS8R likes this.
  10. dasa2

    dasa2 Member

    Joined:
    Feb 5, 2003
    Messages:
    998
    Location:
    Northern VIC
    Could be a BIOS problem so it may be worth trying a step higher or lower in frequency but for two sticks of RAM a few things stand out although I am not sure they would make such a large difference.

    For PULS8R see if these settings make a difference.
    I think these first two values better suit two single rank sticks while your current settings suit 4 single or two dual rank.
    tWRRD should be 1-2
    SCL could be 2-3

    A few other things that may help.
    tWR 12
    tCWL may handle 12
    ProcOdt ~28-32 may work
     
    PULS8R and micsway123 like this.
  11. PULS8R

    PULS8R Member

    Joined:
    Apr 1, 2003
    Messages:
    665
    Location:
    3029
    Thanks I shall try these settings out on my 3800CL15 1T stable overclock as per below.



    upload_2021-8-16_12-46-4.png
     
  12. dasa2

    dasa2 Member

    Joined:
    Feb 5, 2003
    Messages:
    998
    Location:
    Northern VIC
    I think tWR should be tRTP x2 from memory so if it is stable at 6 like it was in your last screen shot then 6 and 12 would be good but since this one is 8 then tWR 14 is correct unless I have it wrong.
    So hard remembering all these rules\guidelines.
     
  13. PULS8R

    PULS8R Member

    Joined:
    Apr 1, 2003
    Messages:
    665
    Location:
    3029
    Haha yeh, I'm slowly learning them, should write them down somewhere, I can never remember the tRFC one.

    Changing a few of those settings worked and got my latency down a bit and read speeds up, but also slightly unstable so a no go on that. I've left 3800 and am concentrating on 4000 now anyway, hope you don't mind me reaching out if I need assistance. I don't know why, but I can't get my latency down, even when I bench at mid 55ns normally booted, diagnostic mode doesn't lower it by much anyway if at all. Not sure how you and other guys are getting 50! Must be the memory controller or something.
     
  14. OP
    OP
    M1ck1

    M1ck1 Member

    Joined:
    May 2, 2018
    Messages:
    575
    Location:
    Toowoomba Qld
    hi mate
    at 4000 I had trouble with whea errors, error ID 19 and stability, maybe IMC, who knows dropped frequency to 3933 and the whea errors are gone. OCCT is good at finding them and even tells their ID . saves searching through event viewer. I usually try TM5 memory test 20cycles as Dasa2 suggested then 30min of OCCT stability test if no errors good enough for me, OCCT puts a lot of stress on your cpu as it runs the memory test.
     
    Last edited: Aug 16, 2021
  15. OP
    OP
    M1ck1

    M1ck1 Member

    Joined:
    May 2, 2018
    Messages:
    575
    Location:
    Toowoomba Qld
    Hi Is there a book or just random snippets around the forums I,m very grateful for said snippets but sometimes some timings I don't understand like AddCmdSetupand its brothers only ever seen a few numbers that people got stability from, you mentioned 56, some other guy had to set it to 6, I just plucked 50 out of my hat and it worked, I didn't use 56 as you used it for 4x8 lol
     
    Last edited: Aug 16, 2021
  16. micsway123

    micsway123 Member

    Joined:
    Aug 7, 2020
    Messages:
    752
    Location:
    Melbourne 3015
    this guide is handy I've found:

    "AddrCmdDrv or AddrCmdDrvStr(en): Address/Command Drive Strength/Impedance. The resistance on the following CPU (or memory controller?) pins: address, RAS, CAS, WE, bank, parity. Theoretically, then, raising it can increase the stability of tight timings."

    https://www.reddit.com/r/overclocking/comments/ahs5a2/demystifying_memory_overclocking_on_ryzen_oc/

    but yea agree, not much info on this one
     
  17. Cynide

    Cynide Member

    Joined:
    Sep 17, 2001
    Messages:
    236
    Location:
    Melb
    Curious, does anyone know what these would do at standard 1.2v?
    I’m more interested at getting more performance with less heat/power than absolute performance with more power.
     
  18. dasa2

    dasa2 Member

    Joined:
    Feb 5, 2003
    Messages:
    998
    Location:
    Northern VIC
    There is a bunch of extremely thick books apparently but all I know is just snippets from around forums (some of which from people that are reading said books) and a few guides.
    AddCmdSetup 48 also worked for me but 56 improved stability further.
    56 is out of sync hence the performance hit from it but the gains for me seem to be more than worth it.

    Extra V to the RAM only increases power use by ~1-3w per DIMM depending on if it is single or dual rank and just how far over 1.5v you go.
    Zen 3 power use jumps by ~7w idle & load just by setting a RAM speed over 3200.
     
    Last edited: Aug 17, 2021
    M1ck1 likes this.
  19. PULS8R

    PULS8R Member

    Joined:
    Apr 1, 2003
    Messages:
    665
    Location:
    3029
    Hey Mick, I'm finding 4000 very stable on mobo/cpu combo. I had WHEA errors that I put down to a bad overclock, so now when testing mem I lock in 4600MHz @ 1.2v so my cpu doesn't influence the results, I'm tempted to drop the frequency further to rule out anything but my WHEA errors disappeared after that. I log them using HWINFO, not sure if this is the same as OCCT, I'll give it a go too..

    I have given up tryin to get GDM disabled @ 1T at anything over 3800, I'm trying 4000 GDM off 2T and tighter timings to see what performance difference I get between that and GDM enabled 1T since GDM limits the tcl to even numbers.

    The formulas are great, if you can go that low, I've found often you have to make up your own values just get it stable and go with a higher number whihc can degrade performance but increase stability. There's a trade off, so it's a fine balance.

    I'll try 3200 C14 on standard 1.2v, this could be nice. Or let me know what you'd like to run? I haven't found any problems at higher voltages though, even running 1.55v the ram doesn't get that hot. I have a watt meter so happy to test W efficiency too.
     
  20. OCMunkee

    OCMunkee Member

    Joined:
    Jun 28, 2004
    Messages:
    1,787
    Location:
    Melbourne, VIC
    FYI - both HWInfo and OCCT will just be capturing WHEA errors being logged via Windows system/error logging so should be the same.

    I still get WHEA errors when running stable memory/fabric clocks (3866/1933), it doesn't seem to impact performance or stability. My theory is that the errors are related to an onboard device (e.g. SATA controller) that is enabled but isn't actually being used.
    Some of the formulas are absolute minimums (e.g. TRAS must be at least tRCD + TRTP), but some are just guidance/suggestions for starting points. Obviously you need to do your own tuning & testing to find the sweet spot between high freq / low timing and stability/performance.

    It will depend on your workload too, some workloads are more latency sensitive, some just want bandwidth. This is worth knowing as some timings (e.g. tRFC) will have a tipping point where lower latency starts to hurt bandwidth and vice versa.

    I like integralfx's guide here: https://github.com/integralfx/MemTestHelper/blob/oc-guide/DDR4 OC Guide.md
    as well as the Reddit thread here: https://www.reddit.com/r/overclocking/comments/ahs5a2/demystifying_memory_overclocking_on_ryzen_oc/

    ::EDIT::

    M1ck1 FYI
     
    Last edited: Aug 17, 2021
    PULS8R and micsway123 like this.

Share This Page

Advertisement: