1usmus Custom Power Plan for Ryzen 3000 Zen 2 Processors

[hzy4]

Did you achieve advertised boost clock of 4600mhz single core, before this power mod?

 

[Ubersonic]

Hi. Why I cant reach advertised 4600 on my 3900x, even with this power plan it is?

View attachment 136024

ASUS MB CH7 1.0.0.3ABBA
CPPC = ON
CPPC Preferred Cores = OFF

That's odd, you should be able to get 4.6GHz on multiple cores (mine does 4.6 on six and 4.4 on the other six and I'm on an ASUS board too). Have you tried monitoring with HWMonitor instead of HWiNFO?

 

[shinarae]

Hi. Why I cant reach advertised 4600 on my 3900x, even with this power plan it is?

ASUS MB CH7 1.0.0.3ABBA
CPPC = ON
CPPC Preferred Cores = OFF

Could be your cooling. I'm running on strix X570-e 1.0.0.3ABBA so the bios you're on should be good.

 

[ORLY]

Windows 10 v1909 (installed from scratch from an MSDN image).
AGESA 1.0.0.3ABBA, couldn't find the "CPPC Preferred Cores " setting as well as those 2 last "insignificant" ones, may be that's why it doesn't work as I expected.

Ryzen 3600, CCX1 contains the fastest core overall, but Windows keeps using the fastest core on CCX0 and the second fastest core on CCX0, ignoring the fastest core of the whole CPU. Tested it with the Cinebench R20.060 single core test. It may be interesting that Windows 10 v1809 (LTSC) usually behaved the same when playing CS:GO - it only used the CCX0 fastest cores.

Even HWiNFO knows which core is the fastest overall and marks it as #1, but Windows still prefers the first CCX, although the best core is on the second CCX.

May be AGESA 1.0.0.4B will fix that, hopefully.

 

[mat9v]

Because 4600MHz is way too high for these silicons. Generally they capable for 4200-4300MHz max, anything is over out of real performance gain.

Right and that's why I happily use my PC at 4.6Ghz (CCD0). What ZEN2 should have is AVX1/2 offset so it would lower multiplier automatically when AVX2 load is detected. Beside that 4.6Ghz is no problem, not for stability, not for temperature. AVX2 is also a problem but not for stability but for heat density - you just can't cool it down without extreme measures above certain frequency (4.3Ghz for very good LC). Your last sentence is curious - are you suggesting that performance is not scaling with clock increase? Well, at least in tasks that do not use AVX it does, linearly at that. AVX2 of course would not work at that frequency due to heat.

 

[Ubersonic]

Because 4600MHz is way too high for these silicons. Generally they capable for 4200-4300MHz max, anything is over out of real performance gain.

Right and that's why I happily use my PC at 4.6Ghz (CCD0). What ZEN2 should have is AVX1/2 offset so it would lower multiplier automatically when AVX2 load is detected. Beside that 4.6Ghz is no problem, not for stability, not for temperature. AVX2 is also a problem but not for stability but for heat density - you just can't cool it down without extreme measures above certain frequency (4.3Ghz for very good LC). Your last sentence is curious - are you suggesting that performance is not scaling with clock increase? Well, at least in tasks that do not use AVX it does, linearly at that. AVX2 of course would not work at that frequency due to heat.

Don't listen to Dyatlov, his monitoring software maxes out at 3.6Ghz.

 

[zstoichev]

Thanks all for the replies.

Did you achieve advertised boost clock of 4600mhz single core, before this power mod?

No, never. The max I see is 4525, both HWInfo64 and Aida64. I think OpenHWMonitors showed 4550 maybe 1-2 times, very rarely. Maybe this should count by AMD as 4600. I don't think so.

That's odd, you should be able to get 4.6GHz on multiple cores (mine does 4.6 on six and 4.4 on the other six and I'm on an ASUS board too). Have you tried monitoring with HWMonitor instead of HWiNFO?

Lucky you. Manually I can do 4300 @ 1,3V on all cores, and CCD0 = 4400; CCD1 = 4300 @ 1,35V. Didn't tried more.

Could be your cooling. I'm running on strix X570-e 1.0.0.3ABBA so the bios you're on should be good.

I am using custom EK WB cooling, should be fine. CPU never goes above 75'C on 4300 all cores, or 85'C on 4400 as said above. Idle at 45. Enabling PBO does nothing.

=> My point it that on stock settings I never ever saw 4600. I can see some of you hit 4650. Not sure if I am missing something or its bad silicon?

 

[shinarae]

I am using custom EK WB cooling, should be fine. CPU never goes above 75'C on 4300 all cores, or 85'C on 4400 as said above. Idle at 45. Enabling PBO does nothing.

=> My point it that on stock settings I never ever saw 4600. I can see some of you hit 4650. Not sure if I am missing something or its bad silicon?

That idle sounds quite high. Mine is 35 or so. What temperature and frequency are you maxing out on Ryzen Master when you run R15 single core with antivirus and other unnecessary background/monitoring processes turned off?

 

[zstoichev]

You are right, when I stop all background processes it idles at 35-ish

Ran some tests with R15 single core. For reference Multi Core R15 score results I get are like 3215.
Average temp on PrefCode Off is 58, and with PrefCode On is 65. You can see max temp difference of 7-8 degrees.
Single core R15 score is 207 when PrefCode Off vs 201 when On. So, better scores and thermals with Off setting.

 

[necrophyte]

i dont have amd cpus, however ive been playing around with the hidden power plan options for years due to obvious improvment potential, so my question is, if somebody knows whether the changes to the preferred cores algorithm will only affect already user-accessible power plan options (HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\PowerSettings), or also some inaccessible hard-coded ones?

if there are only changes to power plan options, one can easily do the same thing prior to upgrading to 1909 already, resp. revert the changes back to pre-1909 after the upgrade, so this is no big deal actually.

 

[q674077444]

Hello God, I am a Chinese lover. I have encountered some problems while using my computer. This makes me very depressed. My 3600 will suddenly get stuck in the running process, and the appearance of Caton is very high. High, the result is a large current sound (I inserted the sound into the display), can you please God to guide the solution or opinion, thank you very much.

 

[HD64G]

News from the developer with suggestions to solve the deviation in boost clocks from PC to PC.

https://twitter.com/i/web/status/1194264084230918144

 

[zstoichev]

Nothing in KB4512941 description speaks about AMD boost fix. Someone tested this?

 

[Arcátko]

Nothing in KB4512941 description speaks about AMD boost fix. Someone tested this?

Since v1909 was made public, that's kinda obsolete workaround iirc.

I've tested removing chipset driver, PC restart, BIOS update to AGESA 1.0.0.4 B, reinstall chipset driver back, PC restart and i still see the Win scheduler using wrong CCX, still bouncing threads between SMT and phys cores but at least now i have low power idle back when using Ryzen Balanced power plan so not bad for me (aka the CPU finally enters idle state and does not cook itself while doing so).

I'm temp bound on the upper end of freq spectrum so i'm not going to see any higher boost clocks with or without this power plan mod.

 

[Wickedt]

After some very cold serious testing, computer was rebooted on the power plan changes, HW reset before running Cinebench R15 single core bench.
So with a very cold day here were i live, i decided to run some tests with 1usmus Power Plan and Ryzen Balanced Power plan. As outdoor temperatures are very cold here, @-6C. I opened the sliding doors and pointed the fan at my open computer for cooling.

Temps never went above 52C while running Cinebench R15 single core bench. 1usmus power plan, CR15 was 209, with Ryzen power plan, score was 201, so thats an 8 point jump.

Temperature definitely affects the CPU performance and boosting.









The point of all this is to see if the 1usmus power plan works as its supposed to, and this proves it does.
If you look at the cores that HW pick as the fastest, 1 and 2 (physical core 2 and core 0 respectfully) , we see a boost for the second choice of 34.2 Mhz, however, look at the core that was picked as the fastest,
it boosted itself 312.7 Mhz to a high of 4624 Mhz! This resulted in the Cine-bench Single core result of 209.

So for anybody with a 3900X or 3950X, this is a no brainer tweak.

PS To anyone concerned, i think the feeling is finally back in my toes after that arctic blast.

 

[zstoichev]

Since v1909 was made public, that's kinda obsolete workaround iirc.

Checked Win Update history, so I have it already:
>> Successfully installed on ‎9.‎9.‎2019

The point of all this is to see if the 1usmus power plan works as its supposed to, and this proves it does.
If you look at the cores that HW pick as the fastest, 1 and 2, we see a boost for the second choice of 34.2 Mhz, however, look at the core that was picked as the fastest,
it boosted itself 312.7 Mhz to a high of 4624 Mhz!

So for anybody with a 3900X or 3950X, this is a no brainer tweak.

This is with Preferred Core On or Off?

 

[Wickedt]

Preferred core on, basically all settings 1usmus recommended, except cool & quiet and PPC adjustment.
These settings are not available on ASRock X570 Taichi.

Test System Specifications:

• Ryzen 9 3900X
• NZXT X62 AIO
• ASRock X570 Taichi (BIOS 2.46, AGESA 1.0.0.4В)
• G.Skill Trident Z Royal Silver DDR4-3600 C16 dual-channel (@ 3733 CL15)
• Windows 10 64-bit 1903
• AMD Chipset Driver 1.9.27.1033

EDIT: added system specs.

EDIT2: Here are results with Prefered Core diasbled.





Lower temps (max 42C), but cores all over the place, never just the preferred fastest cores. CB 15 score of 202, as opposed to 209.
10C jump in heat, which is why cooling is so important.

 

[Ubersonic]

Everyone seems to be getting great results with this power plan/settings, but when I copy them it drops my CB R20 scores from 7421/528 to 7203/527

I don't understand :S

 

[Wickedt]

Everyone seems to be getting great results with this power plan/settings, but when I copy them it drops my CB R20 scores from 7421/528 to 7203/527

I don't understand :S

Are you using Bios 1.0.0.4B?
Also its for 3900X and 3950X soon to come out.

 

[Ubersonic]

Are you using Bios 1.0.0.4B?

Yup.

 

[Wickedt]

Motherboard related maybe? I also un installed the latest amd chipset drivers, re i9nstalled chipset drivers, then changed it from ryzen power plan to 1usmus and rebooted, then tested

 

[biffzinker]

I have the B450 Tomahawk (not max) and i cant find the cppc, all the others options are there but the cppc are not, any help?

The CPPC Preferred Cores option showed up in the newest beta bios update for AMD ComboPI1.0.0.4 Patch B. The CPPC option is in CPU Features at the bottom.

Ryzen Owners Zen Garden

You wrote the chips degrade fast. Do you have info about it? Oh That was someone elses response (jism) not mine ...but im looking into it

www.techpowerup.com

 

[rckidyt]

When opening the .bat file I get the following:

******************************** * 1usmus Ryzen Power Plan v1.0 * ******************************** https://www.techpowerup.com/review/1usmus-custom-power-plan-for-ryzen-3000-zen-2-processors/ Installing... The file could not be found. The power plan has been installed, please activate it via "power options" - "additional plans" Press any key to continue . . .

 

[Wickedt]

Unzip it first to a temp folder on the desktop, run the bat file as admin

On another note, Bios 2.50 is out for ASRock X570 Taichi

ASRock X570 Taichi

Supports AMD AM4 Socket Ryzen™ 2000, 3000, 4000 G-Series, 5000 and 5000 G-Series Desktop Processors; Intel Wi-Fi 6 802.11ax (2.4Gbps) + BT 5.2; Supports DDR4 4666+ (OC); 3 PCIe 4.0 x16, 2 PCIe 4.0 x1; NVIDIA NVLink™, Quad SLI™, AMD 3-Way CrossFireX™; 7.1 CH HD Audio (Realtek ALC1220 Audio...

www.asrock.com

 

[ORLY]

R5 3600, 2 CCX - 0 and 1, the best core unfortunately is on CCX1, I've never seen it go 4200 MHz, moreover - the fastest core on CCX0 has higher turbo frequency than the core marked as the fastest of the whole CPU.
Windows 1909, 1.0.0.3ABBA - it loads the first CCX (CCX0) cores first, only after all 3 cores (w/o SMT) are loaded - it loads the fastest CPU core which is on CCX1. And it starts using SMT only after there are more threads than there are unloaded cores left, for me it's when I run Cinebench R20 with more than 6 threads.

If using CCX1 first drops overall performance - then it makes sense to use CCX0 first. And then this custom power plan only adds a bit of frequency, the rest stays the same visually, I've not benchmarked it, only checked how it loads cores/threads in Task Manager.