1usmus Custom Power Plan for Ryzen 3000 Zen 2 Processors

[HwGeek]

Who is going to bech CS Go ? ;-)

 

[Duvar]

Guys do we need (1.0.0.4b) for this to work?

 

[Anthr4X]

Maybe I'm missing something here, but for the record, I observe no difference/improvement in Cinebench R15 and Cinebench R20 (multi and single) between this and the default High performance power plan. But I also can't find the CPPC (I did find Preferred Cores), AMD Cool'n'Quiet and PPC Adjustment options in my bios. Disabled PBO.

My setup:

• Ryzen 7 3700X
• Corsair H110i GT 280mm
• Gigabyte Aorus X570 Elite (BIOS F10a, AGESA 1.0.0.4В)
• Windows 10 64-bit 1903
• AMD Chipset Driver 1.9.27.1033

//edit: Or is it only the way it schedules and assigns cores that changes with the power plan? No synthetic benchmark improvements?

 

[R-T-B]

We all know the W10 scheduler is trash.

More correctly, the windows scheduler is trash.

 

[TechCat]

Lets remember this shows most improvement over 8 cores muti-chiplet CPUs Both this and 1004B increase efficiency whie lowering temps and will be improved going foward MS Windows is the culprit here and has dropped the ball and had plenty of time, has the resources so has no excuse

 

[xxquaaxx]

Happy user

 

[AndreiD]

R5 3600, performance is mostly the same but I did notice a small improvement in AIDA64 latency for some reason, guess the thread isn't hopped around by the W10 scheduler?
Pretty happy with this overall, before I had PSS & C-States disabled in BIOS and I was using the Ryzen High Performance plan to eliminate some random micro-stutter in games. Haven't tried any games yet with 1usmus' power plan, but so far so good.

 

[Weshya]

Just created an account to thank 1usmus for this wonderful fix and recommended ZEN2 BIOS settings. After switching to the recommended BIOS settings on my Tomahawk MAX B450 (version 3.4 Agesa 1.0.0.4B) and applying the power plan, I was able to notice max boost clock (4.2) across 4 cores out of 6 , on my Ryzen 3600, and for a more sustained time. Finally i get the performance that i was initially promised by AMD.

P.S : Maybe we should create a petition, to claim 1usmus's immediate hiring in AMD

 

[everythingj]

two questions:

does this plan need to be run 1.0.0.3 or 1.0.0.4? Or do either work?

and also, if i'm on a 3900x and was already seeing 4550 sporadically on 4 cores, is that about all that can be expected? i tried applying this hoping it would have those cores hit 4600 but it's not having any effect (on 1.0.0.4)

ironically on 1.0.0.3 i had one core that could hit 4600 but the 1.0.0.4 update has let more cores boost higher so i've kept it applied

 

[Ubersonic]

I'm really confused here, I have the 3900X and my "stock" clocks (with the bundled Wraith cooler) are way better than those in review, does that mean my results with this new power plan would be too? And why are my clocks/scores better? :S

*EDIT*

Ahh, just noticed in the review, the comparison is made vs the bog standard Windows Balanced power scheme, not the Ryzen Balanced optimised power scheme installed by the chip-set drivers.

 

[EarthDog]

I'm still wondering how much this really improves things past pbo and stock (with ryzen pp)...I'm not convinced the effort is worth the result from what we've seen.

 

[silkstone]

I'm still only getting a maximum 1-core boost to 4.1 and all core boost to 3.9 on my R5 3600

 

[Stag]

3800x up 25Mhz in all core tests.Single core boost same as before 4550+.

THANKS

 

[Xzibit]

This is going to confuse a lot because not all Motherboard vendors enable these options and they might be tied to certain features depending on vendor.

 

[Cidious]

Thanks 1usmus for your dedication to the community!

Personally I'd have to say that differences are hard to spot on the 3800X on my system but peak boosts are a bit more evident across cores.

I think I got a few more single core points on R20 but within margin of error and the threads still bouncing around. Although they bounced around on the same core.

Used advised bios settings and PBO enabled and disabled. No difference.

In gaming cores won't run 4500+

but small peak speeds raised slightly across cores that before maybe 1 core boosted 4550 on max peaks and 1 to 4525 and the rest all below 4500. Now I get a few more cores hitting those 4550 speeds only with peaks. But I feel it doesn't make a difference in real performance. Aida Latency is the same. I ran LatencyMon to check about system Latency, couldn't spot much difference there either.

Temps are the same. No decrease in power usage also.

I might have to include that I'm already on Windows 10 19H2 1909 for quite a while. Agesa 1.0.0.4 B and I felt ABBA boosted better for me.

Overall more cores seem to hit their Max boost clocks but nothing shocking. Remind you, I'm on a 3800X.

I'll do some more testing of course. Later this week I'll also rip out the H80i V2 AIO and replace it for a 360 rad to unleash the full potential of the processor and see if that is what's holding it back. Or maybe I just have poor silicon.


UPDATE:
Checking and testing a bit more I find that CPU-Z single core test did improve by more than 10 points. So there is definitely something going on here and not for the worse. System feels smooth and responsive. I'll give it a few more days of testing.

 

[DeathtoGnomes]

@1usmus I Dont have a ryzen rig but I use custom power plans, I Thought maybe you could incorporate this into your research and maybe add some options to this awesome work you did.



Here is the link to the article: Chose Power Plan context menu

 

[mahoney]

Maybe I'm missing something here, but for the record, I observe no difference/improvement in Cinebench R15 and Cinebench R20 (multi and single) between this and the default High performance power plan. But I also can't find the CPPC (I did find Preferred Cores), AMD Cool'n'Quiet and PPC Adjustment options in my bios. Disabled PBO.

My setup:

• Ryzen 7 3700X
• Corsair H110i GT 280mm
• Gigabyte Aorus X570 Elite (BIOS F10a, AGESA 1.0.0.4В)
• Windows 10 64-bit 1903
• AMD Chipset Driver 1.9.27.1033

//edit: Or is it only the way it schedules and assigns cores that changes with the power plan? No synthetic benchmark improvements?

Same with my friends rig. Thought we were doing something wrong but it's the same performance as before (Latest agesa f10a - Gigabyte Aorus pro, 3700x)

 

[rchiwawa]

Using a 3900x and an Asus Crosshair VII motherboard I am seeing a 75MHz increase in my all core clock speed while running a folding@home on the 22 of 24 threads. Seems like a winner to me.

EDIT: This is for the same coolant temp under said load

 

[quadibloc]

The article didn't make clear if this was doing something risky and unsupported to the AMD hardware, or it was only a fix to stupid behavior of Windows 10 that did not involve risking pushing the CPU beyond its limits. While parts of the article seemed to hint at the latter, it also stated that the power plans for AMD processors, instead of being written by Microsoft, were supplied by AMD. I think this means that the custom power plan simply works around bad algorithms that were written by Microsoft. In any case, I learned not to use the "High Performance" plan.

 

[eidairaman1]

The article didn't make clear if this was doing something risky and unsupported to the AMD hardware, or it was only a fix to stupid behavior of Windows 10 that did not involve risking pushing the CPU beyond its limits. While parts of the article seemed to hint at the latter, it also stated that the power plans for AMD processors, instead of being written by Microsoft, were supplied by AMD. I think this means that the custom power plan simply works around bad algorithms that were written by Microsoft. In any case, I learned not to use the "High Performance" plan.

Yes Windows 10 sucks.

 

[moproblems99]

I still need to get 1.0.0.4 flashed but I already get over 4.6 on my 3900x on five cores 4.6 on a sixth. Not sure of the averages but I'll do some brief before and Afters.

 

[wkdzel]

3700X on an MSI X570-A Pro.

I have a baseline on PassMark from 2019-09-05 to compare with (AGESA 1.0.0.3ABB)

Comparing to the latest BIOS I have access to today (AGESA 1.0.0.4B) plus the 1usmus power plan I find that my single core performance tends to be just a few points higher and it does, in fact, tend to keep heavy single threaded workloads on the best core (monitored using Ryzen Master) however it does seem to have transient moments where it tries to move it to a different core and then goes right back to the best core. However all the other tests tend to trend 1-2% lower that my baseline however that is independent from the power plan (I also checked the default AMD power plan) so apparently I lost a bit of performance from 1003abb to 1004b. I benched with both PBO set to disabled and set to Automatic and I saw a slightly better performance with PBO set to automatic but the difference was within 1% so, could just be a fluke. I rant the tests multiple times and it did tend to average higher with PBO on auto but it's a bit too close to call. Honestly I doubt you'd feel the difference in game but I can confirm that with 1usmus's power profile it WANTS to keep the heaviest load on your best core, despite the fact that it will occasionally try to move it for a second and then puts it right back on the best core. (C03 is my best core and it occasionally tried to move it to C01, then right back to C03, every time) where as the default AMD power plan tends to toss the single core test all over the place.

I hope this input helps, thanks for the work!

 

[wolf]

I need to Flash 1.0.0.4 B stable bios and try this out, My gigabyte X570 I (ITX) doesn't want to let PBO work properly at all (wont exceed standard 88w limit) so my best clocks have been in default mode, if this increases efficiency I should see marginally higher clocks within that envelope.

 

[Bwaze]

1.0.0.4B didn't do anything over 1.0.0.3ABBA on my Gigabyte Aorus X570 Elite and 3900X, it lowered the max. boost a bit (from 4.650 to 4.575), but all the results stayed the same (within 1%) - processor doesn't do any work at max boost anyhow.

Tried 1usmus bios settings an power plan, no difference. Tried to push PBO overclocking, minimal increase in scores but massive amount of extra heat.

A bit underwhelming for a supposed massive bios overhaul...

 

[Sashleycat]

Hi, I am really interested in this, especially since I do use a lot of 1T software alongside the heavily threaded ones and the Windows 10 scheduler seems intent on putting the load on the weakest cores in the system. Unfortunately, I tried to use this power plan and it didn't affect anything. CPUz Single-thread uses two threads apparently, and one isn't even on the same CCX. The xEdit software for Fallout 4 uses just on thread on one core, and even with the power plan it still uses the weakest core, sometimes losing 100 MHz of performance which is quite significant.

I am on 1.0.0.3 ABBA and my motherboard is Asrock X570M Pro4, my CPU is Ryzen 7 3700X. Everything is at stock.

However, I couldn't find the:

• CPPC
• CPPC Preferred Cores

Options in the bios. They are simply not there, unless I missed them. So maybe that is why it is not working for me. Any advice would be appreciated. Thanks