The AMD Ryzen All In One Tread /Overclocking/Memory Speeds & Timings/Tweaking/Cooling

Discussion in 'Processors and motherboards AMD' started by chispy, Feb 22, 2017.

Thread Status:
Not open for further replies.
  1. OnnA

    OnnA Ancient Guru

    Messages:
    17,852
    Likes Received:
    6,743
    GPU:
    TiTan RTX Ampere UV
    windows's scheduler has no issues at all with ryzen -

    [​IMG]

    Yes and im a CONAN Barbarian :bang:
     
  2. Dekaohtoura

    Dekaohtoura Master Guru

    Messages:
    382
    Likes Received:
    56
    GPU:
    XFX 6700XT SWFT
  3. Jagman

    Jagman Ancient Guru

    Messages:
    2,264
    Likes Received:
    328
    GPU:
    16GB RX6800
    Good showing by Ryzen 7 against the 7700K I'd say. Still looking forward to Ryzen 5 :)

    @chispy: Can you gain anything extra by overclocking the FSB?
     
  4. chispy

    chispy Ancient Guru

    Messages:
    9,979
    Likes Received:
    2,693
    GPU:
    RTX 4090
    Yes , indeed you gain quite a bit in the motherboards with an external clock generator that allows FSB overclocking. So far there are only 4 Motherboards that have external clock generator - Asus Crosshair Vl Hero , ASRock Taichi , ASRock Fatal1ty X370 Gaming K4 and Gigabyte K7.
     

  5. Jagman

    Jagman Ancient Guru

    Messages:
    2,264
    Likes Received:
    328
    GPU:
    16GB RX6800
    Thanks chispy. I used to like overclocking the FSB on my old Athlons and Phenoms.
     
  6. eclap

    eclap Banned

    Messages:
    31,468
    Likes Received:
    4
    GPU:
    Palit GR 1080 2000/11000
    Overclocking the FSB is the only way to get some DDR4 running at 3200mhz or above atm, from what I've seen.
     
  7. chispy

    chispy Ancient Guru

    Messages:
    9,979
    Likes Received:
    2,693
    GPU:
    RTX 4090
    That's true and absolutely correct my friend , yes. In fact the only way i could overclock my memory to DDR4 3600Mhz and beyond is with FSB overclocking on the Asus CH VI Hero.
     
  8. eclap

    eclap Banned

    Messages:
    31,468
    Likes Received:
    4
    GPU:
    Palit GR 1080 2000/11000
    I take it there's ratios for PCI-E slots etc when overclocking the FSB? I haven't overclocked the FSB in any of my rigs in years, it's really old school :)
     
    Last edited: Mar 13, 2017
  9. chispy

    chispy Ancient Guru

    Messages:
    9,979
    Likes Received:
    2,693
    GPU:
    RTX 4090
    Yes there is.
     
  10. chispy

    chispy Ancient Guru

    Messages:
    9,979
    Likes Received:
    2,693
    GPU:
    RTX 4090
    ** Updates for today March 12 , 2017**

    New Bios for Biostar motherboards:
    ---------------------------------------

    BIOSTAR X370GT5 - X37AG308.BSS , 2017-03-08 , Improved Memory Compatibility
    BIOSTAR B350GT5 - B35AG308.BSS , 2017-03-08 , Improved Memory Compatibility
     

  11. jura11

    jura11 Guest

    Messages:
    2,640
    Likes Received:
    707
    GPU:
    RTX 3090 NvLink
    Hi Angelo

    Can you check with DPC with DPC Latency Checker ?

    Thanks,Jura
     
  12. chispy

    chispy Ancient Guru

    Messages:
    9,979
    Likes Received:
    2,693
    GPU:
    RTX 4090
    Sure will do later , i will post the results and findings here.
     
  13. jura11

    jura11 Guest

    Messages:
    2,640
    Likes Received:
    707
    GPU:
    RTX 3090 NvLink
    Thanks Angelo :)

    Because with Win10 I'm getting crazy DPC with my current X99

    Thanks,Jura
     
  14. chispy

    chispy Ancient Guru

    Messages:
    9,979
    Likes Received:
    2,693
    GPU:
    RTX 4090
    R7 1700x @ 4225Mhz

    Well i got to post this rofl :D

    Just fired up the ASRock Fatal1ty AB350 Gaming K4 and R7 1700x on my bench test custom water cooling loop to test max overclock for a quick cpu-z validation, but once i saw the cpu took the overclock and was semi stable at 4225Mhz 1.55v :eek: i decided to run a quick benchmark of GPUPi CPU Test and it ran the complete benchmark flawless , now i got the number 1 spot for Ryzen 1070x cpu on GPUPi CPU test: http://hwbot.org/submission/3487485_ here are the results:


    Click the spoiler:



    [​IMG]
     
    Last edited: Mar 13, 2017
  15. __hollywood|meo

    __hollywood|meo Ancient Guru

    Messages:
    2,991
    Likes Received:
    139
    GPU:
    6700xt @2.7ghz
    4.225ghz not too shabby :wanker: now asrock needs to get their BIOS sorted

    to elaborate on chispys answer, this is because of the memory dividers available. currently, the selection of ratios is quite limited. i fully expect this to change asap with later BIOS revisions once the mobo manufacturers are finished integrating basic subtiming options.
     

  16. thatguy91

    thatguy91 Guest

    Advanced users only, at least for the testing guinea pig stage.

    I haven't seen people changing these options to see if it helps yet, at least until an official solution if found, if there is indeed and issue. Adjusting some bcdedit options may help, all these should be tried without the useplatformclock entry specified.

    These need to be changed from an elevated command prompt.

    If the OS is muddling the logical processor/thread situation, such as seeing 16 logical processors instead of 8 logical and 8 SMT:
    Code:
    bcdedit.exe /set groupsize 2
    If Windows is seeing the processor as 16 logical processors it will create 8 processor groups, which will mean one logical and one SMT thread per group.
    Note: This would technically be an improper use of it, but worth a try to see if it makes a difference :)

    Alternatively, for the CCX. The CCX is in effect, a NUMA like mode (apparently?)

    Code:
    bcdedit.exe /set groupsize 8
    --> If Windows is seeing the logical and SMT threads incorrectly, it sets each CCX into its own processor group. If it is seeing it correctly, this option will essentially be no change. This is because it groups logical processors, as there are 8 logical processors you are making 1 group (default). If it is seeing 16 logical processors it creates 2 groups, which coincides with the 2 CCX's.

    Code:
    bcdedit.exe /set groupsize 4
    --> If Windows is seeing the logical and SMT threads correctly, it sets the two CCX's into its own processor group of 4 logical processors.

    In addition to the above settings, might need to try this one (off is default)
    Code:
    bcdedit /set groupaware on
    These may or may not affect performance in a positive/negative way:
    Code:
    bcdedit /set nx alwayson
    bcdedit /set tscsyncpolicy enhanced
    The first option shouldn't affect performance, but in the past some have claimed it does. The tscsyncpolicy option could help, but likely won't.

    All this is worth a try though. I haven't built my system yet, so if someone could try this out :). A reboot is a must after each change.

    Could someone test these in this order? Leave the previous settings enabled after each test, the groupsize options will replace the previous test setting.

    Try the settings in this order, reboot after each one and benchmark (including a game benchmark):
    bcdedit /set nx alwayson
    Reboot and test

    bcdedit /set tscsyncpolicy enhanced
    Reboot and test

    bcdedit.exe /set groupsize 2
    Reboot and test

    Either one of these two should in theory be the 'correct' one all else being equal:
    bcdedit.exe /set groupsize 8
    Should give the same results as straight after the tsyncpolicy test (probably..., unless Windows is stuffing things up like some people suggest). If this does improve performance, it suggests something is off with Windows in that it is either seeing the logical threads incorrectly, or it is setting the NUMA correctly for the CCX's and that impedes performance?...

    bcdedit.exe /set groupsize 4
    Sets each CCX into its own NUMA group if it sees the logical processors correctly.

    With the groupsize set, if more than one group is created, the following should be potentially beneficial:
    Code:
    bcdedit /set groupaware on
    Before someone states that this won't do anything, it needs to be tested. If everything is working perfectly it shouldnt't improve performance, but the whole point of the testing is seeing if something isn't, as is suspected. Thefefore, it can only be known if it helps is if someone tests each one of these out with a synthetic and game benchmark in a game that appears to perform more poorly than it should. The more testing of course, the more accurate the results.

    Would be intereting to see the findings. Please note that these are advanced settings, changing the groupsize may result in unexpected behvaviour .That is, if they are actually applied and Windows doesn't override them, again which won't be known until tested.

    Probably ideal testing with a set overclock to avoid discrepancies cause by turbo clock etc. You may need to set the overclock in Windows when testing the groupsize options in case Windows sets it back to default clocks. I noticed this when testing on my current processor (which already I knew would be of zero benefit, just to see how Windows detects it).

    Setting the groupsize manually may result in default clocks, so needs adjusting in Windows. Also it may result in only (for instance) half the logical processors being detected or whatnot. However, it's all about testing and seeing what works or what is beneficial, and if any are beneficial it proves that Windows is not working as expected (however lack of benefit does not mean Windows is workign properly!).
     
    Last edited by a moderator: Mar 13, 2017
  17. OnnA

    OnnA Ancient Guru

    Messages:
    17,852
    Likes Received:
    6,743
    GPU:
    TiTan RTX Ampere UV
    When you get positive results in CCX matter, please show us how to make a .BAT file to speed some things up :nerd:

    THX

    Im waiting for CPU now, MOBO & RAM is in tha house :banana:

    UPD.
    [​IMG]
     
    Last edited: Mar 13, 2017
  18. thatguy91

    thatguy91 Guest

    Yes, will be interesting to see if performance improve. I am still on my i5-3570K, so no benefit, I only did it on mine to see if there was any erroneous behaviour.
     
  19. ddelamare

    ddelamare Guest

    Messages:
    224
    Likes Received:
    5
    GPU:
    Inno3D GTX 1070 X4
    Let me read through this first:
    https://msdn.microsoft.com/en-us/windows/hardware/drivers/devtest/boot-parameters-to-test-drivers-for-multiple-processor-group-support

    If i get time tonight, once i understand, i can do some testing and benching
     
    Last edited: Mar 13, 2017
  20. ddelamare

    ddelamare Guest

    Messages:
    224
    Likes Received:
    5
    GPU:
    Inno3D GTX 1070 X4
    Ok, i've ran a stupid amount of benchmarks at each stage, primarily just focuses on Corona as its cpu intensive.

    I added 'groupaware on','tscsyncpolicy enhanced' and 'nx AlwaysOn' and added/removed the different groupsizes for testing:

    Test 1:
    bcdedit /set groupsize 2
    - only two cpu cores/threads showing (Windows), 100% utilization @ 4.03Ghz
    - Corona bench - approx 450000 rays actual

    Test 2:
    bcdedit /set groupsize 4
    - only four cpu cores/threads showing (Windows), 100% utilization @ 4.03Ghz
    - Corona bench - approx 880000 rays actual

    Test 3:
    bcdedit /set groupsize 8
    - only eight cpu cores/threads showing (Windows), 100% utilization @ 3.6Ghz
    - Corona bench - approx 1700000 rays actual

    Test 4:
    bcdedit /set groupsize 16
    - only sixteen cpu cores/threads showing (Windows), 100% utilization @ 3.6Ghz
    - Corona bench - approx 3400000 rays actual

    So cpu performance seems to be scaling correctly based on its ray compute power in Corona.

    TimeSpy with test 4 - is 6441 points, this is compared to a run before the bcdedit changes which is almost the same:
    http://www.3dmark.com/compare/spy/1369401/spy/1365289#

    I see absolutely no difference with any of the settings other than reducing the amounts of cores/threads and the results are within margin of error.

    What i have noticed is that 3DMark is now showing my cpu boost at 4Ghz in the results, previously this was always locked at 3.592Ghz and would not go any higher. This appears to be due to the 'groupaware yes' option. Made no difference to Corona results or Timespy results.

    Running Timespy with windows 'Performance' profile set is giving me almost identical performance when set to 'Balanced'.
    When set to performance, my cpu is sitting at 3.78Ghz the whole time, obviously.
    When set to 'Balanced', my cpu is idling mostly during the test and ramping up to 90%+ in the cpu test/demo. On Balanced, my cpu goes between idle @ 2.2Ghz, then to 3.7Ghz when utilization is high enough with brief blasts @ 4.1Ghz.

    I dont see any point to the game testing as i play at 4k and my cpu is always at ~20% utilization.

    Heres a list of the TimeSpy Runs:
    [​IMG]

    P.s I also had 'useplatformclock true' enabled in bcedit due to the Ryzen HPET thingamajig... also made no difference :p
     
    Last edited: Mar 13, 2017
Thread Status:
Not open for further replies.

Share This Page