• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

1usmus Power Plan for AMD Ryzen - New Developments

@[B]1usmus[/B] I've had my 3900x for a while now and just got around to playing with fclocks and memory timing. With PBO off,3800 ram and fclock @ 1900mhz I saw 539 in the single core cpu-z bench and 8301 in multi core. With your power plan I saw bump in performance and it now score more consistently! Scores hover around 546-547 for single and 8430-8450 now. The system is noticeably more snappy and I also pickup 12 fps in gears 5 too!

I've been reading articles on techpowerup for years but I specifically join this forum to say thank you!

System configuration
3900x - Cooled by Custom loop
EVGA 1080Ti - Cooled by Custom loop
ASROCk Taichi bio v2.50
g-skill royal 3600 2x16gb @ 3800 xmp profile timings
Fclock @ 1900mhz
PBO - disabled
win1909 latests updates
1usmus universal power plan 1.1 : )
Hey would you care to check Neon Noir benchmark? You have almost the same config as I and I can't get it to run faster then 30fps. CPU load is really high on all cores, but they don't heat up and stay below 45C while GPU is loaded below 30%.
 
Hey would you care to check Neon Noir benchmark? You have almost the same config as I and I can't get it to run faster then 30fps. CPU load is really high on all cores, but they don't heat up and stay below 45C while GPU is loaded below 30%.
This is what I'm getting for what its worth... During the bench FPS are between 140-160 and frame times are 7ms. (Cpu temps peaked at 57c gpu load was around 97%) Our setups are very similar even down to the 4k tv + monitor. lol

My gut says start with clean drivers(if you haven't already) then double check your bios settings...


cryraybench.JPG
 
Last edited:
I did :(
I uninstalled Nvidia drivers, did DDU uninstall. Did bios reset and loaded defaults.
I don't really want to reinstall Windows just to check this.
Look at above screenshot - CPU load is stupid high but cores are barely warm - 45C?, GPU load is low - why?
All benchmarks results are way above average on my setup, but this one? down in a gutters.
 
I would check for redundant PBO and Auto OC features. I know the ASrock bios had these features enabled by default in forefront of the bios and nested away in AMD PBO folder. I know Steve from Gamers Nexus had some weird result between different x570 mobo's.
 
Damn, I just on a hunch did the following:
bcdedit /set Useplatformclock No
and restarted
Now this stupid demo is working correctly.
11400p at 1080p high quality 9200p at ultra quality.
 
@1usmus thanks a lot for the new version... you made my day perfect... did see first time 4.650 wow

So for you guys.. did a few test runs... i am so happy....:)

So first the Speed proof... by the way Gigabyte X570 Xtreme F10 Bios updated today... so i did play the hole day...
4650.jpg

so i did a few test and did see that PBO Auto was on so the speed was less... did turn PBO off. see some more tests...

so I did set the setting system power and sleep to performance after and i did get more score...

well did the tests with HWinfo on and then the test record with this off

and sorry for the bad english...i hope you did understand what i try to show and tell
 

Attachments

  • record.png
    record.png
    43.7 KB · Views: 481
  • test normal.png
    test normal.png
    580.6 KB · Views: 388
  • test performance.png
    test performance.png
    568.2 KB · Views: 390
  • memory.jpg
    memory.jpg
    387.2 KB · Views: 366
  • memory1.jpg
    memory1.jpg
    377.8 KB · Views: 318
  • memory2.jpg
    memory2.jpg
    403.2 KB · Views: 316
  • memory3.jpg
    memory3.jpg
    386.4 KB · Views: 297
Last edited:
I have a similar spec'd machine and I'm curious what the PBO setting should be set to as well. I too cannot boost passed 4.2Ghz with the power profile, and my temps hover in the mid-40s under 10% load with my AIO Corsair H115 Platinum for cooling.
  • Ryzen 3900x - haven't touched OC settings outside of the 1usmus instructions, or Ryzen Master settings.
  • x570 Auros Extreme Mobo
  • 64GB (3600mhz XMP profile) Corsair RAM
  • Windows Build 1903
  • F10c BIOS from Gigabyte website
  • Corsair H115 Platinum AIO
  • 1.09.27.1033 chipset drivers,
  • 1usmus 1.1 power settings and bios settings to match. I know for sure I got the following BIOS settings: Global C-state Control, Low Current Idle, and CPPC Preferred Cores to Enabled - I also couldnt find PPC).

Still not quite there yet.. PBO is disabled, all other background processes disabled. Missing anything?? Temps are in the low-mid 50s during C15/C20.
 

Attachments

  • MaxBoostSingleCore.jpg
    MaxBoostSingleCore.jpg
    214 KB · Views: 317
did you upgrade the BIOS to F10 not F10c... that helped me also, its up 2 days i think

and check the settings in start, settings power and sleep or so called i did that to performance also, there was a post on side 3 i think

i did also mod the Ram timings with ram calculator now, but will post that tomorrow... as to late here, need to sleep...

hey by the way Version F10 will have all the setting you need to support 1.1 that is also new with F10 so I think you will boost more also


see its up from the 21th

so hope you fix it, will talk later
 
did you upgrade the BIOS to F10 not F10c... that helped me also, its up 2 days i think

and check the settings in start, settings power and sleep or so called i did that to performance also, there was a post on side 3 i think

i did also mod the Ram timings with ram calculator now, but will post that tomorrow... as to late here, need to sleep...

hey by the way Version F10 will have all the setting you need to support 1.1 that is also new with F10 so I think you will boost more also


see its up from the 21th

so hope you fix it, will talk later

Agesa 1.0.0.4b + F10C. I'm running 64GB of Corsair Dominator Platinum @ 1800Mhz but I haven't done the RAM calc.
 

Attachments

  • Memory.jpg
    Memory.jpg
    58 KB · Views: 274
I'm still not lucky,
windows continues to use the first 2 cores while the second best core is not the core 2 but the 4 in my case.
the core 1 appear to be good, as per hwinfo (not based on the ryzen master tool) but the core 2 is ranked 4, not 2

I'm using the latest windows insider version and my Asus bios has been updated to.
I'm still not able to keep my cores down when running single thread benches. :(
 
@X570-3900X-DE Did you achieve 4650 with any negative voltage offset or just stock? I'm on Aorus Pro Wifi and still experimenting. I'm getting one core at 4600 and very similar CPU-z scores as you. I can definitely get better scores with PBO maxed out and -0.075 offset using 1usmus universal.
 
Read this

ok, I see, the system get the best average...
but wait... its a single core test!
why spreading this to 2 cores?

by the way, I'm using a 3700x and during the R20 single core tests, if I got 4250mhz, I'm (not) happy!!!!
the system dont push the core to its limit (4.4ghz, but also from a % point of view) and still switch too much between the cores
so the temperature is not raising (63deg in average during the single core test)

well... maybe AMD should provide more control on the starting frequency and the frequency / temperature drop behavior. (like we control the fan speed)
 
ok, I see, the system get the best average...
but wait... its a single core test!
why spreading this to 2 cores?

by the way, I'm using a 3700x and during the R20 single core tests, if I got 4250mhz, I'm (not) happy!!!!
the system dont push the core to its limit (4.4ghz, but also from a % point of view) and still switch too much between the cores
so the temperature is not raising (63deg in average during the single core test)

well... maybe AMD should provide more control on the starting frequency and the frequency / temperature drop behavior. (like we control the fan speed)
Similar issue with my 3700x. Freqs have been more stable with this power plan, but doesn’t click over 4.250ghz in single threaded work loads with similar temps Might hit the max 4.4 for a millisecond, just enough for hwinfo to catch it. So, let’s hope something comes along and fixes that
 
@X570-3900X-DE Did you achieve 4650 with any negative voltage offset or just stock? I'm on Aorus Pro Wifi and still experimenting. I'm getting one core at 4600 and very similar CPU-z scores as you. I can definitely get better scores with PBO maxed out and -0.075 offset using 1usmus universal.
Did it with Auto settings did not play with the CPU volts.
Yes i did try with 75 more clocks the other day but was not happy with F10C as Mem OC always crashed...




Agesa 1.0.0.4b + F10C. I'm running 64GB of Corsair Dominator Platinum @ 1800Mhz but I haven't done the RAM calc.
I would try to update the bios, it helped me a lot, hope this will show you also more speed.

Here are the settings I made with the new F10 bios of the Gigabyte Xtreme

so this is with the fast setting from Memory Calculator32GB G.Skill Trident Z Neo DDR4-3600 DIMM CL16 Dual Kit CL16-16-16-36 (F4-3600C16D-32GTZN) at 3800 with low settings, did a few runs this night so 4650-1 is from the night and the rest from now when i just booted up...
 

Attachments

  • IMG_1160.jpeg
    IMG_1160.jpeg
    290.7 KB · Views: 439
  • IMG_1159.jpeg
    IMG_1159.jpeg
    247.6 KB · Views: 334
  • IMG_1158.jpeg
    IMG_1158.jpeg
    396 KB · Views: 353
  • 4650-1.jpg
    4650-1.jpg
    395 KB · Views: 330
  • memory10.jpg
    memory10.jpg
    710.6 KB · Views: 291
  • memory11.jpg
    memory11.jpg
    719.5 KB · Views: 276
  • memory12.jpg
    memory12.jpg
    739.7 KB · Views: 271
  • memory13.jpg
    memory13.jpg
    706.5 KB · Views: 285
  • recordnew.jpg
    recordnew.jpg
    54.5 KB · Views: 305
  • recordnew1.jpg
    recordnew1.jpg
    489.5 KB · Views: 308
ok, I see, the system get the best average...
but wait... its a single core test!
why spreading this to 2 cores?

The argument is that using a pair of cores is better for managing temperatures than running only on one core, and that using the same pair within the same CCX yields better performance than switching between CCXs, even if on the same die (CCD).

Whether any of that is true or not, I'm not certain. For my part, I think we're discussing trivial differences in performance.
 
Last edited:
The argument is that using a pair of cores is better for managing temperatures than running only on one core, and that using the same pair within the same CCX yields better performance than switching between CCXs, even if on the same die (CCD).

Whether any of that is true or not, I'm not certain. For my part, I think we're discussing trivial differences in performance.
right, but this also means there is a lack of control and configuration available to us.
both AMD ans MS should give us advanced tools to manage these type of configuration.
 
right, but this also means there is a lack of control and configuration available to us.
both AMD ans MS should give us advanced tools to manage these type of configuration.
That part is almost entirely in the MS ballpark - only they control scheduler behavior and it should be there that we should control how system schedules threads on different cores. On the other hand it requires a lot ok knowledge to configure that correctly and there would be a lot of ground to misconfigure and curse MS for low performance ;)
So for most users it would be useless/dangerous to ever mess with.
 
ok guys, I conducted additional tests on my Ryzen 3700x by forcing the affinity mask of cinebench R20
and my bench jump from 460 to 488 points!
so letting windows selecting the best 2 cores = 460 points
forcing using the AMD star core only = 488points

the average clock is 4275mhz vs 4250mhz.

conclusion: for a single core selecting 2 cores to spread the temp, is not the best solution, a single thread app = single core, no need for more.
 
ok guys, I conducted additional tests on my Ryzen 3700x by forcing the affinity mask of cinebench R20
and my bench jump from 460 to 488 points!
so letting windows selecting the best 2 cores = 460 points
forcing using the AMD star core only = 488points

the average clock is 4275mhz vs 4250mhz.

conclusion: for a single core selecting 2 cores to spread the temp, is not the best solution, a single thread app = single core, no need for more.

Im just curious the way to set affinity? Is it in the CB R20 settings?
 
Im just curious the way to set affinity? Is it in the CB R20 settings?
no, use the task manager, go in the detail section.
right click the process and set the affinity.
but do it just after starting the benchmark, because apparently there is like a reset of the setup when we start it.
and after that, you'll have a single core running in a perfectly flat line and constant frequency
and no more heat.
so the argument raised by MS may be applicable... but not sure when and with which CPU.
 
hey im using a ryzen 5 3600 with universal power plan by 1usmus
and PBO enabled +200Mhz
i get these results in cpu-z bench is it good ??
 

Attachments

  • cpuzbench.png
    cpuzbench.png
    16.2 KB · Views: 419
Just an update for everyone. I was having a ton of issues, but after installing the latest F10 BIOS on the Auros Elite, all issues appear to be resolved.

Previously my temps were spiking into the 50-60's and all the cores were firing off with just basic bg tasks, discord, chrome open, but now after the F10 BIOS + the appropriate settings from @1usmus instructions, I'm getting temps in the high 30's, with over half of my cores asleep. To also mention, windows and all my programs started up significantly faster then before as well.

1574544114760.png


Thank you so much!!! @1usmus This is awesome!
 
yes F10 did add the functions we needed to make it fit, did you make CPU-z test? would be cool to see what you get in benchmark.
 
@1usmus thanks a lot for the new version... you made my day perfect... did see first time 4.650 wow

So for you guys.. did a few test runs... i am so happy....:)

So first the Speed proof... by the way Gigabyte X570 Xtreme F10 Bios updated today... so i did play the hole day...
View attachment 137335

so i did a few test and did see that PBO Auto was on so the speed was less... did turn PBO off. see some more tests...

so I did set the setting system power and sleep to performance after and i did get more score...

well did the tests with HWinfo on and then the test record with this off

and sorry for the bad english...i hope you did understand what i try to show and tell
76.4ns is terrible memory latency, that's not running 1:1 with the IF. Even at 3600MHz you should be close to 70ns, if not below with that RAM.
If I were you, I'd go check your UEFI settings.
I'm also pretty certain that the RAM can run at 3800MHz without breaking a sweat, but that takes some more tinkering.

right, but this also means there is a lack of control and configuration available to us.
both AMD ans MS should give us advanced tools to manage these type of configuration.
Why? They don't have any obligation to do so.
It's nice that you and maybe another 0.0001% of their market wants those tools, but it ain't gunna happen.
Most people would just break things tinkering with things like that and who gets the blame then? AMD and Microsoft of course, so that's a big no go as an official tool.
 
Back
Top