DDR4 Registered vs Unbuffered (overclocking)

Discussion in 'Memory' started by cyclone3d, Feb 10, 2019.

  1. cyclone3d

    cyclone3d [H]ardForum Junkie

    Messages:
    12,741
    Joined:
    Aug 16, 2004
    So.. moving up to an X99 system from an X79 system.

    The RAM I currently have is some Samsung 2133 8G registered modules.

    With a few quick settings, I was easily able to get it to boot at 2666. I'm guessing some timings (just changed the main ones and upped the voltage from 1.2 to 1.3) are holding me back from getting faster or maybe it is the fact that they are registered.

    Has anybody here had any good experience getting registered DDR4 to run faster than 2666?

    What about latency and throughput?

    Was trying to look up stuff but not much info is really out there. Supposedly Registered memory will have lower latency than regular unbuffered when running in multi-channel setups but I don't have any other RAM to compare to.. and I would want the same exact timings to verify this claim.

    I would like to see the full gamut of timings for some Samsung based DDR4 that is running higher than 2666. 2800, 3000, 3200, or and higher would be ideal.

    Should I just go ahead and pick up some DDR4 3200 or 3600 while it is on sale for the next few days?

    That way I could compare and it would also give me a spare set to have just in case.
     
  2. dasa

    dasa Limp Gawd

    Messages:
    278
    Joined:
    Feb 19, 2012
    X99 doesn't handle high RAM speeds all that well anyway so even if you did buy a 3200 kit you may have trouble getting it to run that speed.
    Maybe stick with what you have at 2666 and just try tighten up the timings to reduce its latency since you don't really need the extra bandwidth provided by the higher frequency when you already have quad channel.

    What V are you feeding the RAM?
     
  3. cyclone3d

    cyclone3d [H]ardForum Junkie

    Messages:
    12,741
    Joined:
    Aug 16, 2004
    It is currently at 1.3v. Bumping up to 1.35v didn't seem to help but I didn't mess with any other voltages either.

    I had to loosen up the timings to 16-17-17-38 to get it to 2666.

    One review of the faster RAM I am looking at shows it being able to run CAS 13 at 2666. Looks like it should do CAS 14 at 3000.
     
  4. cyclone3d

    cyclone3d [H]ardForum Junkie

    Messages:
    12,741
    Joined:
    Aug 16, 2004
    Well.. trying to find an actual guide for DDR4 overclocking and not finding much useful stuff....

    Anyway, it looks like the board (ASUS X99 PRO) is upping the voltages a lot from stock when just upping the RAM speed so I am setting them all manual and going from there.

    This is RAM without an XMP profile so it is not a lame XMP profile causing the voltage issues.

    System Agent voltage was being auto set to 1.715.. bleh.
     
  5. cyclone3d

    cyclone3d [H]ardForum Junkie

    Messages:
    12,741
    Joined:
    Aug 16, 2004
    Well.. pretty sure that the 2666 wall I am hitting is because this RAM is registered.

    I have almost all the voltages back to stock and it still works at 2666.

    RAM - 1.20
    System Agent 0.872
    VCCIO - 1.10 - stock is 1.05, but it was auto setting it to 1.25 which is way too high.

    Voltage and timing adjustments will not even let it POST at 2800. Just boot loops with codes B0 and B1 repeating.
     
  6. cyclone3d

    cyclone3d [H]ardForum Junkie

    Messages:
    12,741
    Joined:
    Aug 16, 2004
    I ended up ordering 4 sticks of G.Skill RipJaws DDR4 3200. Thought about getting the 3600, but from what I could find the timings are not able to be tightened up as much when running at lower speeds when compared to the 3200 sticks.

    I'll post back with results once I get the system set up and tested with both types of RAM since apparently I am the only one that has actually tried to overclock registered DDR4.
     
  7. dasa

    dasa Limp Gawd

    Messages:
    278
    Joined:
    Feb 19, 2012
  8. cyclone3d

    cyclone3d [H]ardForum Junkie

    Messages:
    12,741
    Joined:
    Aug 16, 2004
    Yeah, but this is on X99, and from my understanding, 3200 is kinda pushing it on X99. From the review on it I saw, I should be able to tighten the timings to at least CAS15.. if my CPU IMC can even handle 3200. I am using 4x8GB sticks, so not really expecting to be able to run much over 3000.
    The CPU is an E5-1650 v3... so Haswell-E.

    The lower latency kits are also priced super high.. over twice the price just to get CAS14 instead of CAS16 3200.

    The ones I got are also supposed to be dual ranked so that should help with latency and throughput a bit compared to single ranked.

    Thanks for the links.
     
  9. dasa

    dasa Limp Gawd

    Messages:
    278
    Joined:
    Feb 19, 2012
    And also make it a fair bit harder to hit 3200MHz
    Let us know how you go.
     
  10. pitingres

    pitingres n00b

    Messages:
    47
    Joined:
    Jul 25, 2018
    I think 3200Mhz would be a real feat, since that's equivalent to DDR4-6400. Oh, you meant MT/s...
     
  11. cyclone3d

    cyclone3d [H]ardForum Junkie

    Messages:
    12,741
    Joined:
    Aug 16, 2004
    Well, X.M.P 3200 just gave all kinds of errors in memtest86+ no matter what settings I changed and so I set it to manual and worked my way up.

    Up to 2666, it worked fine at stock 2133 timings and voltage.

    Couldn't get it to POST at 2800 or 2933 even at the timings and voltage for 3200.

    Set it to 3200 and it came back to life.

    Currently have it tightened up to 15-17-17-36 1T at 3200. Doesn't seem to like tighter primary timings than that at 3200.

    Working on overclocking the CPU. Currently have it at 4.2Ghz with cache at 4.0Ghz.

    Only using memtest86 (memtest86+ won't work with SMP and hasn't been updated since 2013) for testing right now. I don't even have this in a case yet as it is going to be upgrading my main rig.
     
  12. cyclone3d

    cyclone3d [H]ardForum Junkie

    Messages:
    12,741
    Joined:
    Aug 16, 2004
    Up to 4.4Ghz on the CPU and 4.0Ghz on the cache.

    Doing testing in Windows now, but this CPU seems to be a really good sample so far.

    I'm going to have to verify if the G.Skill RAM I have is dual ranked or not. The page I was looking at said it was, but it is only single sided.