Ryzen 2 Single Channel Vs Dual Channel + Best Ram Settings

Discussion in 'AMD Processors' started by gerardfraser, Sep 14, 2018.

  1. gerardfraser

    gerardfraser Gawd

    Messages:
    655
    Joined:
    Feb 23, 2009
    Summary Chart Instead of looking at video
    Desktop-Screenshot-2018-12-08-11-11-27-03.png

    Ryzen 2 Single Channel Vs Dual Channel with normal 3200Mhz timings + Best Ram Settings for my Ryzen 2 System with tight timings, in some games at 2560x1440 Ultra settings.

    To see if there is a difference when gaming at ultra settings with high resolution.

    Ryzen 2 CPU's in combination with X470/X450 boards should all be able to reach 3400Mhz on the memory.

    I tested from 3200Mhz to 3600Mhz but anything past 3466 was about same results,so I left out above 3466Mhz.

    I found the sweet spot for FPS on my Ryzen system and that was 3466Mhz with tight timings.

    A couple weird results ,could be just Nvidia Shadowplay anomaly.

    Sorry no testings at lower resolution or low/medium setting.
    There are plenty of others who have shared there findings with the lower settings and lower resolutions.

    COMPUTER USED
    ♦ CPU - AMD 2600X With MasterLiquid Lite ML240L RGB AIO (Fans 55%)
    ♦ GPU - Nvidia GTX 1080 Ti
    ♦ RAM - G.SKILL Ripjaws V 16GB DDR4 at 3200mhz (F4-3200C16S-16GVK) (1x16)Timing 16-18-18-18
    ♦ RAM - G.Skill Trident Z 16GB DDR4 at 3200mhz (F4-4000C18D-16GTZ) (2x8) Timing 16-18-18-18
    ♦ RAM - G.Skill Trident Z 16GB DDR4 at 3466mhz (F4-4000C18D-16GTZ) (2x8) Timing 14-15-15-39
    ♦ Mobo - MSI X470 - Gaming Plus
    ♦ SSD - M.2 2280 WD Blue 3D NAND 500GB
    ♦ DSP - Dell S2417DG G-Sync 165 Hz Gaming Monitor 2560 x 1440
    ♦ PSU - Antec High Current Pro 1200W

    VIDEO INFORMATION
    ► FPS Monitoring : MSI Afterburner/RTSS
    ► Gameplay Recorder : Nvidia Shadowplay
    ► Edit Videos :VSDC Free Video Editor http://www.videosoftdev.com/

     

    Attached Files:

    Last edited: Dec 8, 2018 at 11:17 AM
    Pieter3dnow likes this.
  2. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    5,982
    Joined:
    Jul 29, 2009
    Command rate at 1t or 2t ?

    Anyway nice benchmarks :) . Did you dig up some of those specialized settings that are floating around on overclock.net?
     
  3. gerardfraser

    gerardfraser Gawd

    Messages:
    655
    Joined:
    Feb 23, 2009
    Thanks for the reply and comment.
    Ya sorry only showed 3466 command rate,should have made that more clear. Command rates are 1T for all.

    Special settings on the ram,just copied what was on the ram stick (F4-3200C16S-16GVK)(1x16)for the 3200 benches.
    3466mhz (F4-4000C18D-16GTZ) I ran 3400/3466/3600 with 14-14-14-30 timings just got worst/weird results.Best results with 14-15-15-39 for some reason or I just could be crazy.
     
    Pieter3dnow likes this.
  4. evolucion8

    evolucion8 Gawd

    Messages:
    909
    Joined:
    Mar 15, 2006
    With my 1700 at 3.80GHz, I was able to raise the RAM to 3,200MHz at 1T with GearsDown disabled at 14-14-14-28.
     
  5. Formula.350

    Formula.350 Gawd

    Messages:
    1,010
    Joined:
    Sep 30, 2011
    This is going to make me sound like an ass, so I apologize in advance heh
    The timings you used at 3200 end up making your results provide people with no useful data. You've taken a slower speed and equipped it with higher latency, and then pit it against a faster speed with tighter timings. It's completely unfair and we're unable to draw an real conclusions from it beyond "Well... yea, those are kinda the results you'd expect..." :(

    If your kits are Samsung B-Die, you should have no problem getting away with 3200 @ 14-14-14, or 15-15-15 (if AMD is allowing for Odd CAS again with the current AGESAs). Otherwise, since 14-15-15 at 3466 works, no reason to not use that at 3200. Though, at 3200 I'd be trying 12-14-14, or if Odds work then 13-14-14 (I was able to get my 3200 CL15 kit to get into Windows at 13-15-15, but wasn't benchmark-stable).


    Ok I admit that I skipped over your hardware list after reading the initial post, and then viewed the chart... However, since I'm not sure which data is correct I'm leaving the above, just struck, but also as a primer of sorts to the rest of my message...

    Your "Single Channel" tests originally confused me based on the chart because you list it as (1x8) and I wasn't sure why you'd be testing without the same amount of Total RAM... Now after I see in your components list, it may be a 1x16GB stick? If so that does make way more sense and makes the data way more usable! :p As such, I'm assuming that it's the Chart that is incorrectly labeled for Single Channel and that it needs to be (1x16)?


    That being said, I do still have a couple critiques of things I'd personally love to see:
    - First is that it'd be nice to see the 3200 speed have lower timings; however, if that's not possible then I understand entirely as it's a lot of RAM and likely Dual-Rank, so might be a lot to ask of the memory controller to run with tighter timings.

    - Second would be to also include another test at 3200 in Dual Channel, but this time with tighter timings in order to more directly (fairly) compare to the 3466 speed. As it sits, the 3200 test is a good data point as it offers apples to apples between 2x8GB (presumably) Single-Rank in Dual Channel and a 1x16GB Dual-Rank in Single Channel. Alas, as I said in my above struck comment at the beginning, the 3466 test is a bit of a "Well, duh" data point lol

    - Third, but equally important, the Subtimings... Are they the same between the 1x16 and 2x8 runs? They can have a significant impact on performance if they've not been configured to be as equal as possible. This would also go for if you would add another 3200 test on the 2x8 with tighter timings for comparison against the 3466 test. While it's a lot of work to nail them all down, the easiest way would be: for the Single vs Dual @ 3200 run the 2x8GB with the 1x16GB's Subs, and if you add a Tight-3200 2x8 then to run with the 3466's Subs (then if you want, try to tighten the Sub's at 3200 to make it the best comparison it can be)

    - Lastly, and this is a long shot, but if your board offers you the ability to explicitly toggle Single and Dual Channel, that it'd be interesting to also see what 2x8GB in Single Channel is capable of (probably just at 3200)

    EDIT: Now after re-viewing the chart, it's definitely interesting that at times the 1x16 has much better 1% and 0.1% results compared to the 2x8. Also that in some tests the 3466's 1% and 0.1% are so much worse than at 3200, despite the superior speed and primary timings... As such, my conclusion is that the subtimings at 3466 are way out of whack and the cause of the performance anomalies, and also why you found that the even-higher speed results 'sucked' in comparison. In all likelihood you can probably get away with using the 3200 auto Subtimings for 3466, and potentially even for higher speeds, but if anything as a baseline to increase in order to get higher speeds stable.

    Though it could also be a case of instability, where it has to drop frames in those games and causes the lower percentile results to be worse. unless you've already done all the high-level stress testing on the RAM and determined those timings at 3466 to be 100% stable?

    Either way they are interesting results and now has me curious about upgrading my laptop (Carrizo APU) from 1x8GB to 1x16GB (since HP screwed the pooch and didn't offer a second DIMM slot). That is, provided the 1x8GB in here is a Single-Rank stick... I wouldn't at all be surprised if it was a Dual-Rank though :\ Damn, CPU-z reports it as indeed being Dual-Rank, oh well.
     
    Last edited: Sep 29, 2018
  6. gerardfraser

    gerardfraser Gawd

    Messages:
    655
    Joined:
    Feb 23, 2009
    Formula.350 Thanks for taking the time to respond

    Yes sorry made a mistake on listing the 16GB Ram Stick on Chart as 1x8 .Video I got it correct.

    Single stick was (F4-3200C16S-16GVK) (1x16)Timings for this ram are 16-18-18-18


    Well I will try to answer the best as i can ,I was doing manual runs and not canned benchmarks so there are differences in each run with the 0.01/1% with default values on sub timings which I kept the same for what I posted.

    ******Tested with gsync on and all monitoring tools and shadow play on which reduced FPS 5-15FPS and effected 0.01/1% minimums significantly. There are two many variables to get reliable results every time you test a game and it really does not make much of a difference with games. *****

    I did do testing with tight timings on 2133/2667/2933/3200/3400/3466/3600 but I just went with what I posted.I tried to keep it somewhat simple with Regular 3200 timings and 3466 tighter timings.99% of people will never touch RAM timings.

    Results 2133/2667 with tight timings the FPS in games were about the same for 2133/2667,and are about 5 FPS behind in some games and others games got the same FPS as 2933/3200/3400/3466/3600 with tight timings.

    There was too much information and would of been way more confusing than what I posted. I am no reviewer ,someone people are really picky or take results to literal.


    EG Timings:If interested
    2133 CL10-10-10-xx ,
    2667 CL10-12-12-xx ,
    2933 CL-14-14-14-xx ,
    3200 CL-14-14-14-xx .

    Sub Timings:


    Bottom line for my Ryzen system Single Channel and Dual Channel with tight timings and all sub timings adjusted playing ultra settings have about the same FPS in games I tested with ram 2933/3200/3400/3466/3600.

    So no real need for Dual Channel memory.I wasted a lot of time which was fun to get the answer I was looking for with single vs dual with 2560x1440 Ultra Settings.

    If you run games on lower settings and resolutions then Dual Channel does help in some games.
     
    Pieter3dnow and Formula.350 like this.
  7. Formula.350

    Formula.350 Gawd

    Messages:
    1,010
    Joined:
    Sep 30, 2011
    Yep, I'm still surprised at the fact there wasn't as much of a difference between the averages. I expected a lot more variance when going to single channel, which also is why I was interested in how --at the very least-- the 16GB stick would've done with tighter timings :)

    However, any time when you learn something, is never 'time wasted'! This was time well spent in my book, even if I will never directly benefit from the info, I'm appreciative of now possessing the knowledge. (Which is also the reason behind my curiosity of performance with tighter timings; the more the merrier.) :pompous:

    I thank you for sharing this all with us! (y)
     
  8. gerardfraser

    gerardfraser Gawd

    Messages:
    655
    Joined:
    Feb 23, 2009
    Thanks again for taking your time to reply.
    Glad to share and any information is always good information.
     
  9. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,435
    Joined:
    Dec 5, 2001
    Try it on an Asus Mobo, I run mine all day at 3533 14-15-15-14-34-48 1t 60 ohm 1.48v...using gskill 3600 RGB cl16d kit
     
  10. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    5,982
    Joined:
    Jul 29, 2009
    Which motherboard :)
     
  11. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,435
    Joined:
    Dec 5, 2001
    Rog strix x370f... Had four kits they all do around 3466 cl12-14 depending on voltage..had the same results albeit more like 3466 on a gigabyte gax370 gaming k3 also...

    More voltage captain!!
     
  12. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    5,982
    Joined:
    Jul 29, 2009
    1.48V little to high for my taste, you are not running your system 24/7 ?
     
  13. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,435
    Joined:
    Dec 5, 2001
    Haven't had a problem in the 18months+ I've owned it.. I run my CPU at 1.55v also..b die likes voltage
     
  14. Formula.350

    Formula.350 Gawd

    Messages:
    1,010
    Joined:
    Sep 30, 2011
    Just be sure to keep in mind those who use their systems for 3+ years and prefer to have the components in working (stable) order so they can sell them after.

    There's a great human analogy here lol Just because someone likes something doesn't mean they should do it every day. Some people like drugs, some like alcohol, but taking in excess of either isn't healthy. As such, just because Samsung RAM lets you overclock better with higher voltages (I don't think it's specifically the B-Die equipped sticks), doesn't mean it will for extended periods. Similarly, just because some people's components are capable of it, with minimal circuit pathway degradation, doesn't mean everyone's are of equal quality.

    Second verse, same as the first, with regards to the CPU. :p

    Just sayin', is all. Particularly with the price of RAM, but the CPUs aren't any cheaper either. I'd be really bummed if my RAM or CPU were needing replacing, even if G.Skill were to warranty it, I'd still be without my gaming rig for awhile! lol I've tried to push my RAM, but I was definitely cringing while trying for moar powa with >1.45V on the DIMMs, and that's even with me being well aware of the technically-safe max on B-Die for short air-cooled runs.

    [/2cents]
     
  15. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,435
    Joined:
    Dec 5, 2001
    I too was like that as soon as I decided to prove ryzen is safe above 1.5 it was over...18 month later still running great
     
  16. Formula.350

    Formula.350 Gawd

    Messages:
    1,010
    Joined:
    Sep 30, 2011
    How many systems do you have running at >1.5V, at what clock, with which model CPUs, and what are their daily workloads?
    In all seriousness though... 1 system does not a statistic make. (doing gaming, web surfing, maybe some video editing I assume)
     
  17. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,435
    Joined:
    Dec 5, 2001
    I'm not the only obe running b die
     
  18. Formula.350

    Formula.350 Gawd

    Messages:
    1,010
    Joined:
    Sep 30, 2011
    Yea but? Did I misunderstand you? :S
    I thought you were talking about your Ryzen's core voltage being >1.5V. Thus my previous message, asking all those things to figure out how you'e determined that 1.55V is safe on Ryzen when AMD has said to proceed with caution over 1.45V...

    Or for that matter, why you'd even need 1.55V since the clocks and voltage scaling doesn't yield anything worth the needed voltage after probably 1.47V anyways. I'm assuming you're running a first gen Ryzen, and if you got a golden sample that'll run at 4.2GHz, the heat it must put off must be quite a bit with 1.55V! But if that's how it's been running at for 18+ months (again, I'm speculating since you've not mention speeds), then high praise to you for that feat :D You have more courage than I do! lol (or a fatter wallet)

    Otherwise, I'm not sure what it is you are running at 1.55V *shrug* Your own words have said Ryzen and "CPU ata 1.55v".

    But yes, I also have B-Die RAM.
     
  19. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,435
    Joined:
    Dec 5, 2001
    It kinda happened by accident, it's a 1600x and under xfr it would go up to 1.57...the weird thing is it's a dud at anything below 1.55 and won't run past 3.85...it runs great at 4180 tho but needs that extra voltage bump...it's an early chip as well, bought it two days after debut from jet.com for $202 after the first time buyer 30% discount...
     
  20. Formula.350

    Formula.350 Gawd

    Messages:
    1,010
    Joined:
    Sep 30, 2011
    Yea, AMD mentioned that under XFR control it will/can run under a much higher voltage, but that it shouldn't be taken as the chip's ability, as it's applying that voltage for just a brief moment under a load.
    That did remind me that they had also mentioned somewhere, and maybe it was a Q&A/AMA with an employee, that it was said that Ryzen is rather self-regulating and has safeguards built in to help ensure things ARE kept under control. If the VCore is set, it can regulate things, but it's when using "Voltage Offset" where it doesn't know that the board is supplying additional voltage, and in those situations it can toast itself


    Anyways, sorry for hijacking your thread, Gerard :sorry:
     
  21. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,435
    Joined:
    Dec 5, 2001
    you didnt hijack it in any way... for a long time i tried and tried to run the cpu under 1.5 but it wouldnt respond whatsoever on a gigabyte board no matter what the offset was... then i got this asus board and thank god i could manually put in hte voltage... but once again the cpu didnt respond over 3.85 unless 1.5v or higher was inputted... so why run 3.9 at 1.5 when i can run 4180 at 1.55?
     
  22. gerardfraser

    gerardfraser Gawd

    Messages:
    655
    Joined:
    Feb 23, 2009
    Well some overclocking on 2600X CPU

    2600X 4350Mhz Cinbench and firestrike runs



    2600X 4250Mhz 1.375v Avg BMW Blender Render


    For those who do not want to look at video

    Cinbench screen with shadowplay off 1513/181
    CB15.jpg
    FIRESTRIKE with shadowplay off
    https://www.3dmark.com/fs/16628258
    Untitled.jpg
     
    Last edited: Oct 9, 2018
  23. Below Ambient

    Below Ambient [H]ard|Gawd

    Messages:
    1,435
    Joined:
    Dec 5, 2001