1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

How to optimally configure Ryzen 3000 Series CPUs

Discussion in 'CPUs' started by Michael Nager, 26 Dec 2019.

  1. Michael Nager

    Associate

    Joined: 26 Dec 2019

    Posts: 7

    I posted this in another thread, but having read a lot of posts here about the Ryzen 3000 series of CPUs and thought that I should post my months long experience and experimentation with this series of CPUs in a thread of its own.

    Here is a very short guide to configuring your Ryzen 3000 Series CPU.

    I have the X570 AURUS XTREME motherboard and had a 3600X in it and now, for over a week, I have a 3950X. Originally I was using a GigaByte X470 AURUS Gaming 7 WiFi rev. 1.1 in conjunction with my 3600X but the following is applicable to both.

    My experience - and this is not just with GigaByte mobos, but with another mobo as well - is that you have two choices, both of which are mutually exclusive.

    1) You can configure your BIOS manually and NEVER use the Ryzen Master utility for anything

    2) Configure your Vcore LLC (Load Line Calibration), in my case turbo, and SOC LLC, in my case high, and getting rid of the logo at boot. Don't configure anything else. After booting into Windows configure the system with Ryzen Master.

    For God's sakes don't try to configure the motherboard with the utility Easy Tune - your system will hate you for it.

    If you try to mix and match the two options you will be headed for a world of pain and frustration.

    The crazy thing is that even if you set the parameters exactly the same in both the BIOS and Ryzen Master, you will have nothing but problems.

    Now as always there are exceptions to the rule, and in my case there are two of them:

    A) When I wanted to overclock my RAM to MCLK and FCLK 1867 from 1800 I had to go into "Settings" and "AMD Overclocking" then, "DDR and Infinity Fabric Freqencies/Timings" portion of the BIOS and set the FCLK value to 1867 there. Ryzen Master will change the MCLK but not the FCLK if you are overclocking your RAM.

    B) With the 3950X being a 16 Core/32 Thread CPU, I don't need SMT for most of the things I do - especially gaming. By turning SMT Off you can take the max clock you achieved previously with SMT On and add 100 MHz clockspeed to each core. The only problem is that Ryzen Master will not actually change from SMT On to Off (or vice versa) as it does with other motherboards I have worked with, so I have to do it manually by going into "Tweaker" then "Advanced CPU Settings" and then set "SMT Mode" to "Disabled.

    I made the mistake of listening to the advice of the Tech YouTubers and of the Tech Media and found out, after a number of weeks of frustration, that as far as the Ryzen 3000 series of CPUs is concerned they haven't got a bloody clue.

    Also ALL motherboards will punt way too much voltage into the Ryzen 3000 series of CPUs. Even Ryzen Master on any of the profiles puts an amount of voltage into the CPUs that will damage them over a relatively short period of time (measured in months, not years).

    To this end, if you want to configure your CPU correctly, then, in Ryzen Master, create a Manual profile and under "Voltage Control" set the "Peak Core(s) Voltage" to 1.3 Volts, and see how far you can clock it stably.

    When you have reached a stable clock at that, then you can try lowering the voltage in increments (in my case it was one increment to 1.29375 Volts) and enjoy a system which will give you good performance, low temps and above all last a long time.

    On my system, using a Noctua NH-U12A air cooler, my system runs stable - and I designate stable as being able to do multiple runs of CineBench R20 back to back - with SMT On (i.e. 16 Cores/32 Threads) at 4.3 GHz and a CineBench R20 score of 10,110 at 1.29375 Volts and with SMT Off (i.e. 16 Cores/16 Threads) it runs at 4.4 GHz for a CineBench score of 7,754 at exactly 1.3 Volts.

    I have ordered and AlphaCool Eisbaer LT360 AIO and I will be interested to see what difference that makes, because running CineBench with SMT On results in a higher temperature than I am personally comfortable with.

    Having now tested the AlphaCool Eisbaer LT360 AIO with three Noctua NF-A12x25 fans instead of the fans that came with it, I will say that the temps are lower than with the NH-U12A but not by all that much. It is however enough to make me feel comfortable with the temps when running CineBench R20 with SMT On and running my 3950X in 12C/32T thread mode at an all-core of 4.3 GHz at 1.29375 Volts.

    To summarise, every Ryzen 3000 series CPU I have been able to test extensively (my own 3600X and 3950X and my friend's 3900X) have their "sweet spot" for performance/voltage at 1.3 Volts. Whatever you achieve with regard to a stable clockspeed at that voltage, if you go even 25 MHz per core higher than that, you will have to punt in a disproportionately higher voltage.

    On my 3600X for instance, just to go from 4.225 GHz at 1.29375 Volts to 4.25 GHz required me to put in 1.35 Volts.

    As with all things, the "Silicon Lottery" has to be taken into account and other people might be able to clock their 3600X or 3950X higher for a given voltage than I can.

    Now to someone who has read this and is going to make the obvious objection with regard to Ryzen 3000 series CPUs, please sit back, consider what you are going to write and you will quickly realise why you are wrong.
     
  2. Woodsta888

    Wise Guy

    Joined: 24 Sep 2013

    Posts: 1,660

    Location: Exmouth, Devon

    So none of the newer BIOS's, AGESA's or chipset drivers made any difference in those months you tested?


    Also you say that all montherboards punt way too much voltage in, including Ryzen master and will make the life of the CPU's months not years...please explain and expand on this please. AMD seem to be wrong along with the MOBO manufacturers.?
     
  3. Minstadave

    Caporegime

    Joined: 8 Jan 2004

    Posts: 26,968

    Location: Rutland

    He comes across a total wally, I wouldn’t humour him.
     
  4. Michael Nager

    Associate

    Joined: 26 Dec 2019

    Posts: 7

    Pretty much, because I remember reading back in the day (before the AMD Marketdroids decided that they needed to be able to tout the GHz and didn't really give a toss how much voltage had to be punted in to achieve it) that TSMC - you know the guys who actually created the 7nm node - stated that the max voltage that could be applied was 1.3 Volts.

    And yes, there have already been reports of people having their Ryzen 3000 CPUs degrade in performance (without having "Overclocked").

    Going from 1.0.0.3ABBA to the original implementation of 1.0.0.4 B on my X570 AURUS XTREME board (F10a) was pretty much of a downgrade in performance.

    By that time I had learned how to configure my Ryzen 3000 so I didn't really suffer from stability issues.

    Since then there has been an F10c followed by an F10 (yes, that didn't make much sense to me either) followed by the BIOS version of 1.0.0.4 B I am now rocking - the version F11, which I have some niggles with, but no real issues.

    If you want to bung in the 1.35 Volts or more that Ryzen Master will hammer your CPU with, be it on ECO mode, Default Mode, PBO Mode or Auto Overclocking Mode - all of which result in a lot more voltage being applied than I do manually configuring my CPU, for about a 10% lower CineBench R20 score and a maximum all-core of about 4.1GHz, then be my guest.

    Just don't come to me complaining when your CPU starts losing attainable clockspeed and/or dies.

    I am sure that there are a lot of people at AMD who know what I know, but who are not talking.

    Then there are those who are talking at AMD and elsewhere - such the baldy headed, and bald faced lying, marketdroid Robert Hallock - who don't have a clue.
     
  5. Woodsta888

    Wise Guy

    Joined: 24 Sep 2013

    Posts: 1,660

    Location: Exmouth, Devon


    How do you 'KNOW'? Put up some evidence otherwise you have as much credibility as JAYS2cents when he put up the 1.3v video. How do you know what people at AMD know?

    And SMT shows during gaming that some games have higher fps with it on and some with it off. Overall, next to no real difference and not really worth bothering with.

    I wouldn't come to you when my CPU dies.... if it does, it does, I'll just buy another one. Haven't seen or heard any increase in failures anywhere yet, please post up your evidence.
     
  6. Panos

    Perma Banned

    Joined: 22 Nov 2009

    Posts: 13,200

    Location: Under the hot sun.

    I will say again thus. Get your ram to 3800C16, set up 1usmus bios and power profile setting and cool the best you want the ryzen 3000 cpus.
    Trying to get CB20 numbers higher, is stupid and pointless. As most usage is light threaded where the CPU's clock to 4.6ghz. (eg gaming).

    If do someone wants to chase benchmark numbers manually OC on BIOS but you will lose the performance on light threaded stuff. Choice is yours.
     
  7. wazza300

    Caporegime

    Joined: 11 Jul 2009

    Posts: 27,066

    Location: BenefitStreetBirmingham

    I cantsee amd releasing cpu's and bios's that pump too much voltage to a point where it degrades the cpu,from what ive seen(i maybe wrong) but 1.4v to 1.45v spikes is the norm?

    As with anything though,lower stable voltage,undervolting wont do any harm at all
     
  8. Panos

    Perma Banned

    Joined: 22 Nov 2009

    Posts: 13,200

    Location: Under the hot sun.

    Voltage spikes are perfectly fine as per numerous AMD tech posts even if these are at factory 1.5v limit.
    What is not good, is forcibly overclocking the CPUs to work over 1.28v at all core boost.
     
  9. Michael Nager

    Associate

    Joined: 26 Dec 2019

    Posts: 7

    I know for a fact that Robert Hallock is full of it all you need to do is watch this video:

    https://www.youtube.com/watch?v=prAaADB9Kck

    You didn't really bother to read what I had posted, did you? (Rhetorical question).

    If you had read what I posted then you would have seen that I wrote:

    Implied in this statement of course is that you will be able to get that increase at the same voltage - in my case it was 4.3 GHz at 1.29375 Volts for SMT On (16C/32T) and 4.4 GHz at 1.3 Volts for SMT Off (16C/16T).

    When you are paying £750 for a CPU then "just buy another one" isn't such an easy option.

    I use CineBench R20 for:

    1) Testing the stability of the clockspeed I have set for the given voltage. If it survives CineBench R20 then it will survive everything else.

    2) Validating that the clockspeed I have set and the voltage actually results in a higher score (for instance setting the voltage to 1.3 Volts and then running at 4.2 GHz after which running at the same voltage but at 4.3 GHz to see if the CineBench score has indeed increased significantly).

    The absolute worst case would be Prime95 with small FFTs which ran stably for over three hours on my 3600X clocked at 4.125 GHz at 1.29375 Volts.
     
  10. Panos

    Perma Banned

    Joined: 22 Nov 2009

    Posts: 13,200

    Location: Under the hot sun.

    You didn't understand my post and reasoning. Your manually overclocks are counter productive, restricting the CPUs to perform only on benchmarks well but not on true applications when the 4.5-4.6 boost when truly needed on light threaded apps like gaming.
     
  11. wazza300

    Caporegime

    Joined: 11 Jul 2009

    Posts: 27,066

    Location: BenefitStreetBirmingham

    Ahh I see
     
  12. MrPotato

    Wise Guy

    Joined: 5 Mar 2017

    Posts: 1,052

    Location: Cambridge

    I'm struggling to see the issue here. Not overclocking here, as I think it's pointless, as let alone, th CPU will boost automatically. If it was like my previous i7 4770k, fair enough, but not the case with the Ryzen.
    Also, the CPU carries a 3 years warranty and I don't remember last time I kept a CPU longer than that.
     
  13. Panos

    Perma Banned

    Joined: 22 Nov 2009

    Posts: 13,200

    Location: Under the hot sun.

    Yep. I play X4, where 2 cores are pushed to 100% and 4-8 cores at 60-80%.
    Those 2 at "stock" speeds, clock to 4550-4600 constantly which is what I need. Contrary to all core boost on something like CB is restricted to 4050-4075.

    If I go and put 4.3-4.4 all core boost, to make the numbers on CB look better, losing that 10% perf on light threaded apps like gaming.
     
  14. Buffalo2102

    Hitman

    Joined: 9 Jun 2004

    Posts: 850

    Personally, I think a lot of the OP is inaccurate, to put it kindly. But I look forward to being proved wrong and getting a brand new CPU under warranty in the next few months.
     
  15. Minstadave

    Caporegime

    Joined: 8 Jan 2004

    Posts: 26,968

    Location: Rutland

    Did you follow the OPs advice to “please sit back, consider what you are going to write and you will quickly realise why you are wrong”?

    OP’s a joker.
     
  16. Space Monkey

    Don

    Joined: 8 Nov 2007

    Posts: 13,941

    Location: Outer Space

    Yeah, I smell fish :p
     
  17. Buffalo2102

    Hitman

    Joined: 9 Jun 2004

    Posts: 850

    So I'm not getting a new CPU?

    Jeez.
     
  18. champion1642

    Gangster

    Joined: 30 Aug 2016

    Posts: 106

    You don't need to "overclock" Ryzen 3000, the CPU does it all itself, the most you probably need to do is enable PBO.

    Extensive testing from a lot of people has shown very little gains from manually OC'ing Ryzen 3000 CPU's.

    The only thing you really need to mess around with is RAM speed and timings, that will give you the most benefit with Ryzen CPU's. Right now have my RAM at 3600Mhz C15 and it is perfect.
     
  19. Panos

    Perma Banned

    Joined: 22 Nov 2009

    Posts: 13,200

    Location: Under the hot sun.

    Not even PBO. Found the best gains I get them if PBO is off, have activated XFR, using preferred cores etc and 1usmus power plan
     
  20. gerardfraser

    Hitman

    Joined: 27 Apr 2014

    Posts: 613

    Op is wrong,but I see you keep posting the 1usmus power plan and it is great for browsing and cinebench ,but did you ever actually test gaming performance on this powerplan compared to any other powerplan.
    Lets say an old game like Sleeping dogs because no one can test every game.

    1usmus power plan consistently loses say 10FPS in gaming compared to other powerplans from my testing.I would post a bunch of test ,no point though.