Ryzen 3000 series how to undervolt/offset properly?

Discussion in 'AMD Processors' started by fakeng, Aug 13, 2019.

  1. fakeng

    fakeng n00b

    Messages:
    13
    Joined:
    Mar 20, 2014
    Any guides out there to set your vcore properly? I have gigabyte aorus x570 pro wifi and There's only auto/normal then some voltages. Setting it to 1.125 and 1.17v (1.225 is auto) produced lower performance as well as higher voltage when the CPU boosts.
     
  2. THUMPer

    THUMPer 2[H]4U

    Messages:
    3,006
    Joined:
    May 6, 2008
    If you drop your voltage your cpu won't boost as high. Leave voltage alone, just make sure BIOS and chipset are up to date and let the CPU do the work.
     
    thebufenator likes this.
  3. Darth Ender

    Darth Ender Limp Gawd

    Messages:
    265
    Joined:
    Oct 11, 2018
    incorrect.

    You can undervolt and still boost to the exact same frequencies you were boosting to stock. I have benchmarks to prove it littered around here in various threads. No loss of performance, no loss in pbo peak frequency...just a loss of temperature. silicone lottery may play a role in the amount you can undervolt, but .1 seems certainly doable. i've run dozens of benchmarks on it and it's been running it for the past day with no issues

    The setting that may be getting overlooked is cpu load line - which compensates for high current draw voltage sag. and cpu vrm frequency, which basically sets the frequency that the digital vrm's run at. They're intentionally set low-ish to produce the least amount of heat needed, and to have room for overclocking ...but well cooled vrm's (which basically all x570's have) can be run at higher frequencies. If you're undervolting, this is basically free since you'll be putting less voltage thru so the additional heat from the frequency bump is offset by the voltage drop. Higher frequencies are needed to keep the cpu stable at lower voltages.

    it's not new to ryzen 3000 series cpus but if you skipped 2000 series, you may have missed it as a suggested step in undervolting.
     
    PliotronX and EvilViking like this.
  4. thebufenator

    thebufenator [H]ard|Gawd

    Messages:
    1,182
    Joined:
    Dec 8, 2004
    Ryzen 3000 will start clock stretching if volts are not high enough. This maintains stability but drops performance. It can give the illusion that the CPU is working fine, but you lose performance.

    Be certain to benchmark before and after undervolting to see if performance degrades. Many cpu's will.

    Also, it would seem your mobo does not support offsets.
     
    THUMPer likes this.
  5. Darth Ender

    Darth Ender Limp Gawd

    Messages:
    265
    Joined:
    Oct 11, 2018
    phoronix has good multiplatform benchmarks that will also log frequency peaks/lows/avgs and temps for the duration of each test. I suggest openbenchmarking.org

    I ran mine before and after. There was no significant change in performance for the 24 tests i ran on the ram and cpu except in the couple that really caused the cpu temps to reach thermal throttling while stock. Those tests ended up being measurably faster with the undervolt due to not hitting thermal throttling.
     
  6. THUMPer

    THUMPer 2[H]4U

    Messages:
    3,006
    Joined:
    May 6, 2008
    yeah makes more sense. I'll have to check mine out again. I re read GN's article of this phenomenon, I should have said clocks appear to be the same but performance drops. They are stating it's a bug where clocks still look like they are normal.
     
    Last edited: Aug 13, 2019
  7. Darth Ender

    Darth Ender Limp Gawd

    Messages:
    265
    Joined:
    Oct 11, 2018
    Ran 24 different tests in Phoronix say otherwise. No loss of performance
     
  8. THUMPer

    THUMPer 2[H]4U

    Messages:
    3,006
    Joined:
    May 6, 2008
    what cpu what board and what offset?
     
  9. RamonGTP

    RamonGTP [H]ardness Supreme

    Messages:
    7,679
    Joined:
    Nov 9, 2005
    There's been other people who have also done test that contradict your findings. Most, if not all the stuff I've read or watched lose performance while maintaining clocks. My own testing included with Cinebench. 1 vs many, i'll go with many
     
  10. Darth Ender

    Darth Ender Limp Gawd

    Messages:
    265
    Joined:
    Oct 11, 2018
    I'd take multiple independent benchmarks over a single one

    cpu: 3900x
    mb: asus pro x570 workstation ace

    pbo set to enable
    cpu load line and other current settings set to 110%
    cpu vrm freq set to 400
    cpu core voltage offset set to -.1

    https://openbenchmarking.org/result/1908139-HV-1908103HV27

    You'll notice the 0.05 uv didn't go as well as the 0.1uv. That was before adjusting the cpu vrm freq. I believe that is at least partly integral to keeping the cpu stable and functional at high frequencies at lower voltages. And why I'm seeing no performance loss but others are.

    though, 0.05 is such a tiny undervolt that most of the numbers are within the range of error I'm sure. Or it could be that the 0.05 undervolt wasn't enough to lower the temp enough to be significant.

    Still, I'm not crashing, I'm not seeing an inability to hit the same frequencies I hit before undervolting. I'm not seeing any loss in performance, just a 10% loss of avg and peak temps.
     
    Last edited: Aug 13, 2019
  11. Darth Ender

    Darth Ender Limp Gawd

    Messages:
    265
    Joined:
    Oct 11, 2018
    comparison between my undervolted computer and someone else's stock. Not apples to apples....but still. 0.1v undervolt and faster than this guy's stock 3900x using completely different benchmarks than the other 24 tests run against myself. I was only monitoring cpufreq of cpu0 so you dont get to see the max pbo across all cores but that gets too messy to view when you're viewing 24 cores. Suffice it to say, the reports about undervolting the 3900x causing performance regression are premature at best.

    https://openbenchmarking.org/result/1908145-HV-1907124HV85
     
  12. thebufenator

    thebufenator [H]ard|Gawd

    Messages:
    1,182
    Joined:
    Dec 8, 2004
    Almost sounds like each CPU can have a different quality of silicon.
     
  13. Darth Ender

    Darth Ender Limp Gawd

    Messages:
    265
    Joined:
    Oct 11, 2018
    Sure,. But what's being passed around isn't people saying they are unlucky. They're saying it can't be done without negative consequences.

    It obviously can be done without negative results.
     
  14. PliotronX

    PliotronX 2[H]4U

    Messages:
    2,060
    Joined:
    Aug 8, 2000
    Has anyone seen specifics on the VID going higher to compensate for much lower vcore? Too much of that can harm the CPU, no?