Post your Ryzen memory speeds!

Discussion in 'AMD Processors' started by bobzdar, Mar 15, 2017.

  1. THUMPer

    THUMPer 2[H]4U

    Messages:
    3,041
    Joined:
    May 6, 2008
    Been tinkering with some settings. I feel pretty good about this. However, I'd like to get the latency down. Any suggestions?

    [​IMG]

    [​IMG]
     
  2. Formula.350

    Formula.350 [H]ard|Gawd

    Messages:
    1,107
    Joined:
    Sep 30, 2011
    IF you're interested... The one we're using, is in "Tools" menu > "Cache and Memory Benchmark".
    Click on "Start Benchmark", wait a minute or two, click "Save" to create an image, and voila! :D
     
  3. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,438
    Joined:
    Dec 5, 2001
  4. Formula.350

    Formula.350 [H]ard|Gawd

    Messages:
    1,107
    Joined:
    Sep 30, 2011
    Even when inputting all the same timings/subtimings from your 3200@3500 into the 3466, it still isn't stable or as fast? Tis indeed curious, if that's the case. However, we do know that there are some peculiarities with >3333 since most are having issues running that. Which, there are no doubt plenty of inaccessible timings that could very well be changing substantially and causing that, too.

    Also, when you said "tCWL auto was 24 but was not stable going any lower so left that alone" what all had you tried? I suspect that the CHVI might be one of the boards Stilt was referring to about being 'broken', but in my own testings I found that, say a setting of 13 would run at 14. So my thinking is that if all you tried was a setting of 23, that may be a cause, as your RTC screeshot shows it at 14 and I'm wondering if it's not set at +10 in the BIOS which leads to values input not being translated as 'whole numbers'.
    *cough* That, or it was just a typo and you meant to put 14... LOL Which, if that's the case, then nevermind! :p

    While I doubt this is the cause, but I'm running on the Anniversary build of 10 (Ver 1607, build 14393.447), so perhaps somehow that could be influencing the memory/cache results on your system?

    For what it's worth, I took a moment to compare our RTCs and noticed a couple huge differences between our configs (though, one can't ignore that you're running 167MHz faster than I am). These are MY settings with Yours written (shoddily) in red. The one with the exclamation point I had found interesting. (I didn't mark any of the others because they're the TurnAround timings and are on Auto, the rest I don't have access TO change them)
    My RTC.png

    Worth noting I think that might help your speeds: tWTRL and tWR being lowered if you can. The GearDownMode, if it helped with your stability then fair enough; however, from my reading of the AMD blog it sounds like it might be better to leave it Enabled. Lastly, just a theory, perhaps your tWRRD (exclamation point) being so low is not providing enough time for other latencies to finish their tasks? After all, you CAN tighten certain ones too much and cause performance to drop.

    Lastly, I did notice your overall AIDA version. I doubt that's the cause of the performance difference though, as in my testing of using the really old version's BenchDLL files in the new one, the results were the same as using the old EXE version. I will agree that having the AGESA readout there is nice, but on my board the string still reads 1.0.0.4a so... would be a bit misleading. (MSI did release a new Beta last night, and visually I saw no changes, so I assume it's all under-the-hood stuff... maybe like AGESA string being updated heh)



    Frankly? Your timings are 'fine', as in normal and expected, from my experience. The only thing I can suggest for lowering them, which would only result in 1ns or maybe 2ns, is overclocking the CPU. Otherwise, your latencies are what I was getting before I did the SCL @ 2T (which you've done).
     
  5. THUMPer

    THUMPer 2[H]4U

    Messages:
    3,041
    Joined:
    May 6, 2008
    Whew, that 1600x is screamin! Nice!
     
  6. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,438
    Joined:
    Dec 5, 2001
    Its running on Auto lol
     
  7. THUMPer

    THUMPer 2[H]4U

    Messages:
    3,041
    Joined:
    May 6, 2008
    even better
     
    Below Ambient likes this.
  8. noko

    noko [H]ardness Supreme

    Messages:
    4,342
    Joined:
    Apr 14, 2010
    Awesome, a lot of those settings are what ASUS XMP which they call D.O.C.P (may have that wrong) set at standard. The variance is huge but I am not sure how significant that is. I guess the best way is to backup windows, create restore point and try some of those numbers you have. ASUS maybe way out to lunch on some of the settings or they have to for whatever reason set some of those values there. I don't know.

    Maybe take them one at a time and run down the results but to ensure each one is stable I can only do a short test and once everything configure the long 12+ stability test. Then again how much real difference would all of this make in anything other than some benchmark numbers? I really don't think that would be much but my curiosity is getting the best of me.

    Thanks for the feedback.
     
  9. Formula.350

    Formula.350 [H]ard|Gawd

    Messages:
    1,107
    Joined:
    Sep 30, 2011
    Something Chew* had told me though, is that with memory timings, different models of board will differ a lot on what they'll be able to accomplish. Some of it I understand the reasoning behind, as traces no doubt make a world of difference. However, he was also saying that timings associated with them (I assume that we can't see/adjust) are also affected a lot differently between vendors. Where one might have a lot more relaxed than the other, which in turn will let you tighten up certain timings way more than what the others are capable.
    Regardless, I still believe in sharing settings since it at least gives people like us something to TRY, even if it doesn't work. It also still can provide a learning experience, not just about DRAM, but also about what our boards and components are capable of.

    I'd love to test out what impact there is on games, but I'm in a slightly unique situation with internet where I can't download anything I want, it's on a 'need' basis. The games I do have, will require me to do some registry editing in order to get Steam to think I've re-downloaded and installed them lol (It's all legal, don't worry. They're my games from my own library, I just don't have the luxury of re-downloading them, so I have to copy them from system to system and import registry dumps)

    Still, there's a few GB/s between our configurations on the RAM side, and a heck of a lot more than Cache side, both of which I can only imagine would benefit SOME games in terms of performance :D
     
    noko likes this.
  10. noko

    noko [H]ardness Supreme

    Messages:
    4,342
    Joined:
    Apr 14, 2010
    Looks good to me, have you tried BCLK OCing? Use 3200 strap and take it up to like 104. If you can set your pcie to Gen2 for any M2 drives and Nvidia cards seem to do OK with Gen 3 if you have then. Not sure about AMD cards. For me BCLK OCing for ram speed is more stable.

    Formula 350:
    Have duplicated your settings (the ones that can be adjusted) except remaining at 3500mhz using 3200 strap and BCLK. So far so good. Some improvement with speeds but no where near what you show. I was amazed that at 3500 I can have 14-14-14-14-36 with tCWL at 14. Maybe MSI or you figure out a good combo. Anyways stress testing now.
     
  11. Nobu

    Nobu 2[H]4U

    Messages:
    3,234
    Joined:
    Jun 7, 2007
    Yeah, I managed some ridiculous timings and speed for my 2800 kit that was almost 100% memtest stable with bclk, but I had to back off due to bus instability. With another motherboard I could probably manage to keep it there--this one lacks the necessary voltage options.
     
  12. Formula.350

    Formula.350 [H]ard|Gawd

    Messages:
    1,107
    Joined:
    Sep 30, 2011
    Yea, I'm beginning to wonder what's real and what isn't... lol I'd have thought the fact that >3333 was being so finicky for me meant something to do with my RAM. Then I decided to try 3333 at my 3200 timings and it was stable! Like, I did not expect that... I'd love to try for lower, but w/o being able to run tCL 13 and tCWL 13, I'm not too hopeful that it'll work if I try it lol

    Maybe I'll flash the new 1.74 Beta BIOS today with its mystery changes... :p
    Also with determining that CLDO is the internal DDR Phy voltage, and that I was having that restart issue with the DRAM Training voltage at 1.5V (seemingly fixed at 1.45V)... maybe 3466 will work :D
     
  13. noko

    noko [H]ardness Supreme

    Messages:
    4,342
    Joined:
    Apr 14, 2010
    Yeah, I would post something but want to make sure it is reliable - just don't have much time to do long tests. With the CH6 I just don't like leaving it unattended to - it might go up in a puff of smoke :LOL:
     
  14. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,438
    Joined:
    Dec 5, 2001
    gigabyte is such a peaceful motherboard...
     
  15. noko

    noko [H]ardness Supreme

    Messages:
    4,342
    Joined:
    Apr 14, 2010
    As in dead? :p
     
    JustReason and Below Ambient like this.
  16. Formula.350

    Formula.350 [H]ard|Gawd

    Messages:
    1,107
    Joined:
    Sep 30, 2011
    Welp, I've pulled down all my images of benchmarks with overclocked CPU :\

    Apparently when speeds are adjusted from outside of the BIOS (like I was doing, using K17TK program), the Real Time Clock issue with Windows 8/10 causes benchmarks to incorrectly report their performance. So my changing of speeds to 3.8GHz inside Windows would result in performance numbers that are incorrect. :cry:

    Here are my real numbers :(
    cachemem-DDR4-3200 14-14-14-30 Moderate Mk.III (Stock Ryzen-Balanced Plan) BIOS v1.73.png

    I am no longer a special snowflake. *sniff*
     
  17. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,438
    Joined:
    Dec 5, 2001
    I wish I could say yes but the thing is a real piece of work... totally drama free since i got it... I can't speak any higher about this board until it gives me issues, but its run 4.1Ghz turbo/3200 since i got it, and the bios does what it should when you take it too far... I'm actually enjoying the cool n quiet and automatic modes, the cpu just flat out plows, but then again i came from the age when auto features were garbage, and enabling cool n quiet just made your system lock up...
     
    noko likes this.
  18. noko

    noko [H]ardness Supreme

    Messages:
    4,342
    Joined:
    Apr 14, 2010
    Except your bios settings work, since I have a few days off I will dive back in to see how far they can go. They are sufficiently different then ASUS settings to be rather interesting.

    Edit:
    Your settings worked for about 96min with MemTest64 while playing music, playing videos/web browsing computer screen went black and then blue screen. Tried a few modifications but they failed even faster.

    Went back to mine and modified a few settings particulary tWRRD and it lasted 4 hours of MemTest64 and other stuff and gave good numbers in Mem&Cache benchmark. Could not post an image here not sure why. These are just numbers unless the better numbers reflect anything substantial in other programs, we maybe just chasing after the wind, so to speak.

    MemTestJune12.jpg

    Ah ha! Edge does not work for dropping images, Chrome does. Still loading stuff into Windows after clean install.
     
    Last edited: Jun 12, 2017
  19. Formula.350

    Formula.350 [H]ard|Gawd

    Messages:
    1,107
    Joined:
    Sep 30, 2011
    Sure they work, mostly I think. :p I've yet to truly vet my system with any semblance of stress testing besides a few minutes with AIDA's Stress Test. However, I'm encountering many instances of it being "pretty stable", but seemingly not enough where and when it really counts.

    For example, the system seems to be pretty stable at 3.8GHz with auto volts (and, just to be clear, I always try to state CPU speed in GHz, and RAM in in MHz), but then yesterday it just randomly crapped out, as well as SOMEtimes will flake out when resuming from Sleep. Also I set CPU to 3.9GHz and still on the 1.35V, which ran for over an hour on AIDA's CPU stress test, pulling 145W package (125W 'actual' according to HWiNFO readings of Volts x Amps = Watts) but only putting out 60C (40C on Tdie). Then ~5mins after turning on the FPU Stress test (while things were still running on CPU Stress), it failed. Interestingly the package consumption had dropped to 133W (I think 'actual' remained the same) and the TEMP rose to 83C (63C Tdie), so who knows if that's the reason or not... I still don't trust these temps being at all accurate as I know my Water2.0 PRO is more than up to the challenge since it's the double-thick radiator.

    Either way, in embarrassing honesty, I'm not being a thorough overclocker and making sure things are stable first. :oops: I let the 'uber-but-false' scores with the increased CPU multi in Windows distract me from my RAM work, so really I need to get back to that before I even start trying to increase CPU and find stability with it.

    As for mine not working for you, as I had mentioned that Chew said, the board designs will result in different timings being possible, as well as the BIOS configuration for hidden values we can't access. One can still use them as a reference, though, for certain. Problem though, becomes a matter of "which timings end up having a relation to each other and should get adjusted at the same time" or if not a matter of being changed at the same time, perhaps one of "which timings should always be within XX ticks from each other".
    In other words, you 'adjusting' mine and things failing faster may be a result of you adjusting Timing X, but should've been adjusting Timing X and Y.

    I'm going to see if maybe Chew will be willing to divulge whether or not there are certain ones like that.

    Anyways I just decided I'd flash Beta 1.74 that was released on the 9th. Will give it a shot, and then I think I'm going to roll back to Beta 1.72 since that was the one time when 3466 worked, as well as tCL with Odd values. Fingers crossed :)
     
  20. os2wiz

    os2wiz Gawd

    Messages:
    543
    Joined:
    Nov 20, 2011

    Formula, I saw your last post on extremesystems.org regarding turning HPET off in bios. But I have never seen this setting in the Titanium bios. Does it go by another name?
     
  21. Formula.350

    Formula.350 [H]ard|Gawd

    Messages:
    1,107
    Joined:
    Sep 30, 2011
    I replied there, but I'll reply here, too...
    They decided to make it hidden since either BIOS v1.20 or v1.30, don't quite remember which.
    Since I mod all the ones I run in order to get BCLK (for the whole whopping 3 MHz options lol), I make it visible along with various other options (tRFC2, tRFC4, I think CLDO_VDDP, and the DRAM "(BIOS)" and "(Training)" voltages which I'm not certain if they are functional inputs or mean to be displays).

    I can upload the latest version, but I make various other settings my own defaults just to save time since the OC Profiles generally won't apply between versions due to (I assume) new settings. It's nothing serious, but it might be annoying lol Like my Boot options I set it so that its boot order is: Non-UEFI HDD, Non-UEFI USB Thumbdrive, Non-UEFI CDRom are all first. The Sleep State LED to Dual-Color so that it does blink (makes it turn off for me), and then SATA 5 and 6 have Hotswap enabled. Just let me know which version. Only recent ones I don't have on-hand are v1.50 and v1.60 releases, but I have the Betas, which I only mention since I believe you run v1.60 Release.
     
  22. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,438
    Joined:
    Dec 5, 2001
    Very nice!!
     
  23. jbc029

    jbc029 Limp Gawd

    Messages:
    469
    Joined:
    Mar 26, 2013
    Update on ASRock Pro4 B350 boards. Bios 2.6.0 is out. With that release, the XMP on my 2x16GB TridentZ kit at 3000MHz (@2933) boots reliably at rated speeds. Max stable OC on my 1700 did falter slightly. Now 3.95, down from 3.975. Pleasantly surprised.

    http://valid.x86.fr/1jfuhe
     
  24. Gman1979

    Gman1979 Gawd

    Messages:
    637
    Joined:
    Sep 21, 2011
    I'm really hoping these RAM issues are squared away by the time we can buy into the Threadripper platform. RAM tweaking and playing with timings were always something I enjoyed trying to get the most out of, even if every generation of RAM tech diminished the returns from doing so (still remember noticing the speed difference when I got SDR RAM to run CAS 2 the first time, god that makes me feel old ffs). It's also the main issue that prevented me from buying into RYZEN on launch. If I'm throwing money into satisfying my geek needs, I'd rather do so on a platform with the quirks worked out.
     
    drescherjm likes this.
  25. noko

    noko [H]ardness Supreme

    Messages:
    4,342
    Joined:
    Apr 14, 2010
    Ryzen is definitely the geeky tweaking time fix turning knobs, switches and watching computer melt before your eyes. Long time having that kind of geek high. CH6 bios 1401 (being tested prior to launch, may or may not be the official bios) is doing better with memory for many and with me. So far 3333 with 14-14-14-34 is stable which was not the case with the previous beta bios's. Having hopes for the 3600+ ram speeds now. First time running TPU MemTest64, music playing, browser open and other stuff and zero stutter - smooth as cold beer.
     
    Below Ambient likes this.
  26. Nobu

    Nobu 2[H]4U

    Messages:
    3,234
    Joined:
    Jun 7, 2007
    Seems like timings affect optimum cldo_vddp voltage, not just freq. Moving cl up or down destabilizes my ram. :/
     
  27. boocake

    boocake Limp Gawd

    Messages:
    166
    Joined:
    Jul 11, 2015
    I can't even seem to get any cldo_vddp settings to stick at all whatsoever.
     
  28. Nobu

    Nobu 2[H]4U

    Messages:
    3,234
    Joined:
    Jun 7, 2007
    937-1000 (mV) is apparently the range to look in. I managed 3267cl16-19-19 a minute ago on my gskill hynix m-die ram with it set to 947 (iirc). It was unstable, but I think that was due to the bclk oc.

    Oh, and you need to power down completely for the setting to stick. My gigabyte board does this automatically (either due to failed oc detection or some other reason, not sure), but you may have to manually do it.
    Edit: 3200cl16-16-16 almost one pass memtest86 complete.
     
    Last edited: Jun 14, 2017
    Geno750 likes this.
  29. Gman1979

    Gman1979 Gawd

    Messages:
    637
    Joined:
    Sep 21, 2011
    My Gigabyte (Intel) motherboard in the rig in my sig exhibits similar behavior after changing only certain ram timings. I've been told it's the UEFI "retraining" certain RAM parameters in response to the settings changes. There's a few second delay between when it reboots and gets dialed back in before starting the rest of the boot process.
     
  30. Nobu

    Nobu 2[H]4U

    Messages:
    3,234
    Joined:
    Jun 7, 2007
    Mine turns off completely (no lights, fans, or anything), then back on fans blazing, then it reboots a few times while training the memory (can tell by debug leds and the fans ramping up and down), until it finally settles and the fans go to appropriate speed for temp (or it fails, powers off one final time, and boots with default ram settings after retraining).

    I say one final time, because with bad ram settings it may power down two or three times before successfully booting, either at what I set or at defaults.
     
  31. Gman1979

    Gman1979 Gawd

    Messages:
    637
    Joined:
    Sep 21, 2011
    That actually describes how it acts perfectly! The bit about known stable, but tweaked settings, seemingly being a craps shoot on booting is my only major gripe with Gigabyte on this board so far.
     
  32. boocake

    boocake Limp Gawd

    Messages:
    166
    Joined:
    Jul 11, 2015
    Hmm, I'll see if I can get a setting to stick by manually powering down. After I set it, I do the 'Save & Exit', and as it's rebooting, do I power it off prior to Windows loading?
     
  33. gigaxtreme1

    gigaxtreme1 2[H]4U

    Messages:
    3,507
    Joined:
    Oct 1, 2002
    Let it run, see if you can get to windows. Check speeds with cpuz if you get to windows
     
  34. Nobu

    Nobu 2[H]4U

    Messages:
    3,234
    Joined:
    Jun 7, 2007
    That should do it. In the thread I linked, The Stilt suggested pressing the reset button after saving, before it boots up again. Don't know if that works or not personally, but you can try both and see.
    gigaxtreme1 Waiting for windows to boot to check is a waste of time, imo. At least in my motherboard firmware, you see the current ram speed on the right immediately after entering the firmware, so I just hold the delete key until the bios loads to see if it was successful, and then select whichever os I want from there (usually my Linux bootloader, to run memtest86 and verify stability)
     
    Last edited: Jun 14, 2017
  35. boocake

    boocake Limp Gawd

    Messages:
    166
    Joined:
    Jul 11, 2015
    Well, this is super weird. I ended up flipping the power switch on my PSU to shutdown after setting it within BIOS (did a save and exit, flipped the switch before booting) and it looked like it stuck when I entered BIOS after powering up from a cold start. However, when I restarted within Windows and went back into BIOS just to see, it's back at 'Auto' again.
     
  36. Nobu

    Nobu 2[H]4U

    Messages:
    3,234
    Joined:
    Jun 7, 2007
    I wouldn't recommend using the PSU switch, as that bypasses any firmware control and may leave you in a bad state (though a cmos reset may help recover in that case). Just hold the power button for 3-5 seconds. ;)

    Having said that, that is really weird.
     
  37. Nobu

    Nobu 2[H]4U

    Messages:
    3,234
    Joined:
    Jun 7, 2007
    Managed to accidentally(!) boot at 3200cl14-15-15-15...was dialing in procodt at 2933 with tight timings, changed multi but forgot timings. Unfortunately memtest unstable (lots of errors), but fun none-the-less. Loosening secondary timings doesn't help, so must be just a bit too tight for these chips.

    Fwiw, procodt is good between 48-80 Ohms on one stick and 48-68.8 Ohms on the other (both set to 60 as that's close as possible to middle on this board). Both Hynix single-rank from a dual-channel kit.
     
  38. boocake

    boocake Limp Gawd

    Messages:
    166
    Joined:
    Jul 11, 2015
    So I actually ended up doing just that - holding down the power button for 3-5 seconds before booting up again. Since this new 1401 beta wasI applied, I get prompted to enter the BIOS with a message indicating that I shut the machine off via the power button and would need to check settings (something of the like). It's at this stage when I'm in the BIOS to check if my setting stuck, and that's where I come across seeing it reset back to 'Auto' for some reason.
     
  39. gigaxtreme1

    gigaxtreme1 2[H]4U

    Messages:
    3,507
    Joined:
    Oct 1, 2002
    Load defaults a couple of times and reboot. Settings seem to have persistence
     
  40. boocake

    boocake Limp Gawd

    Messages:
    166
    Joined:
    Jul 11, 2015
    Will give this a shot tonight when I get home from work!