1usmus Power Plan for AMD Ryzen - New Developments

[Ubersonic]

Exactly the same for me and my 3900x. Highest boost was with agesa 1.0.0.3 abba 4625mhz. Agesa 1.0.0.4b lowered them to max 4550mhz.

I got max 4.65GHz with both, but the 1usmus plan/settings got me to an occasional 4.7GHz, so it seems to vary by chip/board.

 

[Tahna]

What kind of temperature should I be at under load to reach the 4.6 clock speed? Starting to wonder if that's my issue. I'm cooling with an AIO h115i in a NZXT h510 elite case with radiator mounted in the front, fans on the outside toward the glass; bios have the pump at full speed, and when I manually set the fans to 100% my idle temps are around 54-55 C.

Care to throw up some screenshots of your UEFI settings? Just plug in a FAT32 formatted USB drive and hit F12 to grab a screenshot.

Here are my settings. Also a SS of my power plan.

http://imgur.com/a/Hlw2qA1

Do I understand correctly that your processor also does not receive maximum boost in single-threaded loads?

Guys, everyone who needs help, attach to your message a log made using Hwinfo 6.15 in CB15 (single thread).

P.s. System must be in default + there should not be any background activity during logging.

Attached are my logs, I ran a single thread, and all thread log. Let me know If I can provide anything else.

Forum wouldn't accept CSV, so here is a dropbox link: https://www.dropbox.com/s/ukan8egygzpfqye/Tahna Hwinfo Logs.zip?dl=0

I've also added a report on my system from HWiNFO if that helps: https://www.dropbox.com/s/btx8sbln4w50f0n/DESKTOP-8TJHVEE.MHT?dl=0

Thanks for the help!

 

[TKnockers]

Ryzen 3600x.. with this 1.1 universal version power plan in windows 1909 my all core boost and single core boost are lower than with 1.0 version.

 

[mat9v]

3900X on AGESA 1.0.0.4B on Asus X570 Prime Pro
Windows 1909 1usmus universal plan 1.1
Regardless of PBO being disabled or enabled, maximum clock hovers around 4.55Ghz.
Cooling is powerful enough, 420mm Alphacool setup with 6 SilentWings 3 fans, on liquid metal, temps are kept below 60C up to 10 cores of sustained load, even with CB15/20 MT load (24 threads) they never exceed 66C on stock. Anyway, bios have CPPC, and idle power control but not CPPC preferred cores, I suppose it is enabled but hidden.
1 thread load uses core2 (gold one) on CCX0 but using 2 threads see the second thread default to core0 on CCX0 instead of core4 on CCX1 (silver one), increasing thread count to 3 brings core1 to work and further using 4 threads brings core3 to work instead of core4 (silver). So it seems that CPPC does not work correctly.
I suppose that I can't reach 4.6Ghz due to thermal constraints, but I simply can't cool the CPU below 50C, even 1 thread load hovers around 55C, even undervolted by 0.0625 (the lowest stable voltage - offset actually works on this board) I can't get it below 53C. Oh well, on 1.0.0.3 ABBA I could get 4.625Mhz.
Anyway, for now, the best approach for me is per CCX overclocking in BIOS (that actually also works well), I can get CB20 stable at 4.55Ghz(CCX0), 4.50(CCX1), 4.4(CCX2-3) and 1.475V (FIT limit) and gaming stable at 4.6/4.55/4.4Ghz at the same voltage - all due to temperature differences (CB20 MT tops at 93C and gaming rarely jumps over 82C).
I can get 7800p in CB20 out of this configuration - https://drive.google.com/open?id=1s2XPkib51TlRIGxZ3oUo6MyvqGH6ji7R
I can only hope that AMD can get us our 4.6Ghz back with future AGESA version.

 

[1usmus]

What kind of temperature should I be at under load to reach the 4.6 clock speed? Starting to wonder if that's my issue. I'm cooling with an AIO h115i in a NZXT h510 elite case with radiator mounted in the front, fans on the outside toward the glass; bios have the pump at full speed, and when I manually set the fans to 100% my idle temps are around 54-55 C.


Here are my settings. Also a SS of my power plan.

View attachment 137144

http://imgur.com/a/Hlw2qA1

Attached are my logs, I ran a single thread, and all thread log. Let me know If I can provide anything else.

Forum wouldn't accept CSV, so here is a dropbox link: https://www.dropbox.com/s/ukan8egygzpfqye/Tahna Hwinfo Logs.zip?dl=0

I've also added a report on my system from HWiNFO if that helps: https://www.dropbox.com/s/btx8sbln4w50f0n/DESKTOP-8TJHVEE.MHT?dl=0

Thanks for the help!

What i found in your case:

1) Average background activity of your system in this test is 10.8%, which is almost 2.1 times more than normal. That is, your background programs do not allow the processor to work effectively with single-threaded tasks.
2) Average temperature is 64 degrees for a single-thread test. It's a lot. Be prepared for the fact that the processor will already lose 100 MHz due to temperature.
3) Vcore 1.39V , this confirms that there are problems with background activity.

I have a suggestion, re-record the log in safe mode and also make polling time 50ms.

Hi there. See attached log of HWInfo 2.15.3390, logged while running CB R15 using your universal power plan. Screen shots of "About my PC" and "Power Plan" settings here. See this reddit post for details on my hardware config and some other benchmarks I've run on Windows 1903 and 1909 using the .

I'm curious to see what you think. I was seeing 4625 Mhz peak on preferred cores at some point in my past (could have been 1003ABB or 1003ABBA) but now the highest peak I ever see is 4550 Mhz on preferred core.

I'm actually happy with this performance, I think it's great, so I'm not really concerned. This is more "for science" than anything else. I'm also willing to do any testing you'd like, if it helps, on fresh installs of Windows and BIOS reset/flashback.

Thank you for all you do for our community.

Thanks! Your system is fine. I like.
If not difficult, set polling time to 50 ms and make logs for the Ryzen Balanced and Universal profiles.

Hi there. See attached log of HWInfo 2.15.3390, logged while running CB R15 using your universal power plan. Screen shots of "About my PC" and "Power Plan" settings here. See this reddit post for details on my hardware config and some other benchmarks I've run on Windows 1903 and 1909 using the .

I'm curious to see what you think. I was seeing 4625 Mhz peak on preferred cores at some point in my past (could have been 1003ABB or 1003ABBA) but now the highest peak I ever see is 4550 Mhz on preferred core.

I'm actually happy with this performance, I think it's great, so I'm not really concerned. This is more "for science" than anything else. I'm also willing to do any testing you'd like, if it helps, on fresh installs of Windows and BIOS reset/flashback.

Thank you for all you do for our community.

There are some changes in the new SMU that relate to frequency at a certain temperature. It looks like your 58 degrees are the cause. I have a suggestion. In BIOS, the limit throttling change from Auto to 105 degrees.

Please note that this offer is only for this situation and only for this experiment.

I got max 4.65GHz with both, but the 1usmus plan/settings got me to an occasional 4.7GHz, so it seems to vary by chip/board.

Very cool!
Can you make a log?

@1usmus - Yuri, what about the mobile variants of Ryzen?

I for example have the 2700U running Win 10 1909 w/ the AMD chipset drivers, which bring the 'Ryzen Balanced' power plan.

Needless to say, the battery life is far from stellar even under low load. With all fairness, laptop doesn't feel sluggish under any task.

Can something be done for the battery life of the mobile Ryzens while maintaining responsiveness?

I unfortunately do not have a mobile processor for this kind of experimentation, but it seems to me that the main condition is a well-maintained operating system. This will allow the device to work longer, since the cores will not wake up again.

3900X on AGESA 1.0.0.4B on Asus X570 Prime Pro
Windows 1909 1usmus universal plan 1.1
Regardless of PBO being disabled or enabled, maximum clock hovers around 4.55Ghz.
Cooling is powerful enough, 420mm Alphacool setup with 6 SilentWings 3 fans, on liquid metal, temps are kept below 60C up to 10 cores of sustained load, even with CB15/20 MT load (24 threads) they never exceed 66C on stock. Anyway, bios have CPPC, and idle power control but not CPPC preferred cores, I suppose it is enabled but hidden.
1 thread load uses core2 (gold one) on CCX0 but using 2 threads see the second thread default to core0 on CCX0 instead of core4 on CCX1 (silver one), increasing thread count to 3 brings core1 to work and further using 4 threads brings core3 to work instead of core4 (silver). So it seems that CPPC does not work correctly.
I suppose that I can't reach 4.6Ghz due to thermal constraints, but I simply can't cool the CPU below 50C, even 1 thread load hovers around 55C, even undervolted by 0.0625 (the lowest stable voltage - offset actually works on this board) I can't get it below 53C. Oh well, on 1.0.0.3 ABBA I could get 4.625Mhz.
Anyway, for now, the best approach for me is per CCX overclocking in BIOS (that actually also works well), I can get CB20 stable at 4.55Ghz(CCX0), 4.50(CCX1), 4.4(CCX2-3) and 1.475V (FIT limit) and gaming stable at 4.6/4.55/4.4Ghz at the same voltage - all due to temperature differences (CB20 MT tops at 93C and gaming rarely jumps over 82C).
I can get 7800p in CB20 out of this configuration - https://drive.google.com/open?id=1s2XPkib51TlRIGxZ3oUo6MyvqGH6ji7R
I can only hope that AMD can get us our 4.6Ghz back with future AGESA version.

Thanks for your feedback.
I think you're right, at the moment the limit is temperature. I will try to influence AMD in this direction.

 

[mat9v]

@1usmus - I suppose that if I were to use distinct programs to generate load, it may just place the load on different CCXes. What I mean is that default Windows behavior is to place threads from the same program as close to each other as possible so it would fill CCX0 before spawning threads on CCX1 - closeness to common L3 cache would be more important then selecting better quality core that MAY boost higher. I think I will test that if I have some time. Use CB15 ST, CB20 ST, CPU-Z ST, 7-zip ST test... hmmm.

 

[Weshya]

On a side note (but still relevant) i noticed that on my Desktop PC, under Power & Sleep>Performance and energy, there's a slider that can be used to control the power plan in use (?). Since i really don't know if by choosing another preset (other than the default "better performance") it would replace my current power plan (1usmus Universal) , i decided to make a quick test.
I noticed that if i choose "Best performance", it behaves as if I've chosen the "High Performance" power plan, BUT retaining the 1usmus universal power plan benefits.

So i am kind of puzzled as to what this setting has to do on a desktop PC. Is it to fine tune the already chosen power plan under control panel ?
I suspect that the "better performance=balanced" option should be chosen if our current power plan is 1usmus, or not ?

Running Win'10 1909

 

[Danielsweb]

Thanks! Your system is fine. I like.
If not difficult, set polling time to 50 ms and make logs for the Ryzen Balanced and Universal profiles.
There are some changes in the new SMU that relate to frequency at a certain temperature. It looks like your 58 degrees are the cause. I have a suggestion. In BIOS, the limit throttling change from Auto to 105 degrees.

Please note that this offer is only for this situation and only for this experiment.

Thank you for the guidance. I ran three different test pairs (odd tests are 1usmus Universal, even tests are Ryzen Balanced). I zipped up the log (50 ms polling) along with some screen shots of BIOS and Power Plan for easy reference of state of each test:

Tests 1 and 2 -BoostfMaxEn

I found this setting in SMU options and set it to 105. I realized later F probably means frequency and not Fahrenheit. I did the tests, here are the results, but feel free to ignore - EXCEPT, I did see 4625 Mhz peak on best core in CCX0 during these tests! (The only time I saw it...)

Tests 3 and 4 - Default

I decided to run a pair of tests with my default BIOS (note: this is NOT stock). I was just curious.

Tests 5 and 6 - Platform Thermal Throttle Limit

I finally found this setting - it was buried under PBO settings, and not in SMU options. I think I got my intent right (leaving PBO off, while increasing the throttle limit) - but considering it was in the PBO area, I'm wondering if it was all moot.

--
I observed some interesting things, so I'm curious what you find, assuming you even have the time to crawl these logs.

 

[1usmus]

Thank you for the guidance. I ran three different test pairs (odd tests are 1usmus Universal, even tests are Ryzen Balanced). I zipped up the log (50 ms polling) along with some screen shots of BIOS and Power Plan for easy reference of state of each test:

Tests 1 and 2 -BoostfMaxEn

I found this setting in SMU options and set it to 105. I realized later F probably means frequency and not Fahrenheit. I did the tests, here are the results, but feel free to ignore - EXCEPT, I did see 4625 Mhz peak on best core in CCX0 during these tests! (The only time I saw it...)

Tests 3 and 4 - Default

I decided to run a pair of tests with my default BIOS (note: this is NOT stock). I was just curious.

Tests 5 and 6 - Platform Thermal Throttle Limit

I finally found this setting - it was buried under PBO settings, and not in SMU options. I think I got my intent right (leaving PBO off, while increasing the throttle limit) - but considering it was in the PBO area, I'm wondering if it was all moot.

--
I observed some interesting things, so I'm curious what you find, assuming you even have the time to crawl these logs.

Thanks for the test.
On your system, the AMD profile is trying to use the wrong core. I want to draw the attention of users to this nuance.

 

[teeradbacesi]

An interesting article from AMD_Robert on best core selection between Ryzen Master and Windows:

https://www.reddit.com/r/Amd/comments/dzjs7i
Bullet 5 seems to apply here:

"Now we’re at the handoff to the OS. This is where things get more complicated. Windows selects and prioritizes the fastest core in the firmware with an additional criterion that there must be a second core in the same CCX that’s nearly as fast. The scheduler rotates between them to ensure one core isn’t shouldering all the single-threaded work all the time. (This is why you will sometimes see a “one thread” task jump back and forth between two different cores.) Additionally, I believe it’s now widely understood in this community that corralling workloads within a CCX, when possible, is optimal for “Zen 2” performance. Windows 10 May 2019 Update also respects this. So, if Windows is going to pick and use a fastest core, it needs a partner within the same CCX to ensure all the criteria are met for optimal performance. This is the best-performing configuration for 1T and lightly-threaded scenarios. "

 

[1usmus]

On a side note (but still relevant) i noticed that on my Desktop PC, under Power & Sleep>Performance and energy, there's a slider that can be used to control the power plan in use (?). Since i really don't know if by choosing another preset (other than the default "better performance") it would replace my current power plan (1usmus Universal) , i decided to make a quick test.
I noticed that if i choose "Best performance", it behaves as if I've chosen the "High Performance" power plan, BUT retaining the 1usmus universal power plan benefits.

So i am kind of puzzled as to what this setting has to do on a desktop PC. Is it to fine tune the already chosen power plan under control panel ?
I suspect that the "better performance=balanced" option should be chosen if our current power plan is 1usmus, or not ?

Running Win'10 1909
View attachment 137181

I will check what this setting does

 

[Tahna]

What i found in your case:

1) Average background activity of your system in this test is 10.8%, which is almost 2.1 times more than normal. That is, your background programs do not allow the processor to work effectively with single-threaded tasks.
2) Average temperature is 64 degrees for a single-thread test. It's a lot. Be prepared for the fact that the processor will already lose 100 MHz due to temperature.
3) Vcore 1.39V , this confirms that there are problems with background activity.

I have a suggestion, re-record the log in safe mode and also make polling time 50ms.

Thanks 1usmus, but I am having an issue getting hwinfo to even start in safe mode. Is there a work around for this? I tried everything in this thread I found, but the methods failed to work.

In regular boot mode, I've disabled all backround tasks in the windows privacy settings, closed everything I can and this is where my tasks land. Chrome would be closed upon CB15 test, and hwinfo would be the only other program open besides CB15.

 

[Danielsweb]

Thanks for the test.
On your system, the AMD profile is trying to use the wrong core. I want to draw the attention of users to this nuance.

That is very interesting. 1usmus universal power plan it is!

Thanks 1usmus, but I am having an issue getting hwinfo to even start in safe mode. Is there a work around for this? I tried everything in this thread I found, but the methods failed to work.

Hi Tahna. Try going into Task Manager, Startup tab, and disabling everything you see there. Once done, reboot and re-run your tests. This is what I do, and I find it gives me a reasonable amount of headroom to do all my tests. Some things are still running (e.g., Radeon software, Windows active anti-virus protection, Bluetooth driver, etc.) but nothing that takes over the whole machine. Give this a try and attach new logs.

 

[Ubersonic]

when I manually set the fans to 100% my idle temps are around 54-55 C

Something is wrong either with your cooler or your CPU mount/settings. Looking at that it's possible either the cooler isn't making proper contact (or isn't working) or the CPU is running higher voltage than it should.

At 100% fan speed you shouldn't be idling in the 50s, the stock cooler can do better than that at 100%.

 

[WiredTexan]

I will check what this setting does

I found this article on tenforums which sheds some light on this 'feature'. Still confusing. (The Power Mode bar does disappear if we choose the Windows High Performance plan)

Edit:
This Channel9 video provides a good bit of information on this feature, comparing Power Plan vs Power Mode. Power Mode deals with power throttling background apps. I found that with my Gigabyte Aorus Pro Wifi X570, no matter what setting I used only System Interrupts showed in the task manager as having power throttling enabled. So it may be a no-op feature on a desktop. It could also be showing up by mistake as the setting makes little sense on a desktop.

 

[Danielsweb]

"For day-to-day system use, our guidance remains unchanged: install any version of Windows from May 2019 (or newer) and keep it up-to-date, grab the current chipset driver from July 7 onwards, and use the latest BIOS from AGESA 1002 onwards. You will see the expected “fastest core” experience. No additional updates are required. If a user wants to confirm with their motherboard as well to remove all doubt, then you can force-enable the following BIOS settings: Global C-States, CPPC, CPPC Preferred Core(s). These settings are available in the AMD CBS menu of virtually any motherboard. Our official guidance is auto/default ON for these settings, but it’s a step you can take out of an abundance of caution."

What I find interesting about this is that yes, this is their guidance. However, at least based on Hwinfo logs from some limited testing on my system, 1usmus was able to demonstrate that the Ryzen Balanced plan is not using the best core as reported by Windows. So personally, I'm going to go ahead and take that extra step and use the 1usmus universal plan. This won't translate into any dramatic difference in performance, but I'll happily take another 2 or 3 fps in my favorite games when I'm not crunching numbers.

 

[Tahna]

Hi Tahna. Try going into Task Manager, Startup tab, and disabling everything you see there. Once done, reboot and re-run your tests. This is what I do, and I find it gives me a reasonable amount of headroom to do all my tests. Some things are still running (e.g., Radeon software, Windows active anti-virus protection, Bluetooth driver, etc.) but nothing that takes over the whole machine. Give this a try and attach new logs.

I'll give this a try, thanks!

Something is wrong either with your cooler or your CPU mount/settings. Looking at that it's possible either the cooler isn't making proper contact (or isn't working) or the CPU is running higher voltage than it should.

At 100% fan speed you shouldn't be idling in the 50s, the stock cooler can do better than that at 100%.

Thanks for the reply. This is what I was afraid of. I feel confident in the mount, but it's still possible something is set wrong. I read another thread with the same AIO (Corsair H115i) / CPU as me, and corsair replaced the AIO which solved the cooling issue completely for them. I'm going to open up my case panels, put on two more fans, and see if that improves the temps.

@Ubersonic how can I make sure the CPU voltage is correct? I haven't messed with anything voltage wise.

 

[Weshya]

I found this article on tenforums which sheds some light on this 'feature'. Still confusing. (The Power Mode bar does disappear if we choose the Windows High Performance plan)

The article you link refers to laptops, whereas i am talking about desktop PC. That's why i am really puzzled as to what this setting is really doing, i could not find any proper documentation either.
It does not seem to replace the current chosen power plan when selecting a different option, but it seems to be fine tuning the current power plan ? I eagerly wait for 1usmus opinion on that.

 

[WiredTexan]

es a good bit of information on this feature, comparing Power Plan vs Power Mode. Power Mode deals with power throttling background apps. I found that with my Gigabyte Aorus Pro Wifi X570, no matter what setting I used only System Interrupts showed

The article you link refers to laptops, whereas i am talking about desktop PC. That's why i am really puzzled as to what this setting is really doing, i could not find any proper documentation either.
It does not seem to replace the current chosen power plan when selecting a different option, but it seems to be fine tuning the current power plan ? I eagerly wait for 1usmus opinion on that.

Yup, most info on this feature will be geared toward laptops. I linked a Channel9 video that compares power plans vs power mode in my original reply and that page provides links to dev documentation. Power Mode deals with throttling power to background processes/apps. The video shows how to see what apps have power throttling enabled, on my setup only system interrupts show this as enabled no matter what power mode I select. Since a desktop doesn't run off a battery I doubt this setting has any affect at all and should not be showing up. Power Plans, however, make sense because they can save you money! Putting monitors, hard drives, etc to sleep when not in use is a good thing, (for the environment and the wallet), whether on a laptop, tablet or desktop.

 

[Weshya]

Yup, most info on this feature will be geared toward laptops. I linked a Channel9 video that compares power plans vs power mode in my original reply and that page provides links to dev documentation. Power Mode deals with throttling power to background processes/apps. The video shows how to see what apps have power throttling enabled, on my setup only system interrupts show this as enabled no matter what power mode I select. Since a desktop doesn't run off a battery I doubt this setting has any affect at all and should not be showing up. Power Plans, however, make sense because they can save you money! Putting monitors, hard drives, etc to sleep when not in use is a good thing, (for the environment and the wallet), whether on a laptop, tablet or desktop.

I see on Reddit that Robert got involved in that matter, and is also surprised as to what this setting is doing on a desktop PC. Your findings of throttling background applications seems to agree with my observations in my quick test. So i guess we should leave the default " better performance" so it does not interfere with AMD's or 1usmus power plans.

 

[1usmus]

I see on Reddit that Robert got involved in that matter, and is also surprised as to what this setting is doing on a desktop PC. Your findings of throttling background applications seems to agree with my observations in my quick test. So i guess we should leave the default " better performance" so it does not interfere with AMD's or 1usmus power plans.

Maybe I'm not attentive. Please describe your observations

P.s. duty cycle disabled by default in Ryzen Balanced

 

[Weshya]

What i have noticed for sure is that setting the slider to "Best Performance" power mode in "Power & Sleep>Performance and Energy" apparently changes the minimum processor state to 100%, because i can see that the min clocks are much higher when my PC is idle. It does that while keeping the already chosen power plan (in control panel) and in my case, it's your universal custom one. So it does not seem to replace the existing power plan, but rather only changes the aggressiveness/snappiness of the cpu on the chosen power plan ?

Edit : "Best performance" changes the idle CPU voltage from ~1.0 to ~1.45 here also

 

[zstoichev]

AMD Clarifies "Best Cores" vs "Preferred Cores" Discrepancies For Ryzen CPUs

www.anandtech.com

 

[Tahna]

Hi Tahna. Try going into Task Manager, Startup tab, and disabling everything you see there. Once done, reboot and re-run your tests. This is what I do, and I find it gives me a reasonable amount of headroom to do all my tests. Some things are still running (e.g., Radeon software, Windows active anti-virus protection, Bluetooth driver, etc.) but nothing that takes over the whole machine. Give this a try and attach new logs.

Well I think the BG tasks were my issue... I applied your method of disabling all the startup applications, re ran CB15 single core with only Hwinfo open and I saw two cores max out at 4591.7!

@1usmus Here is my updated log with polling rate of 50 and all the bg tasks closed. I think i'm performing well and good now, but can you please verify? https://www.dropbox.com/s/94e7aibskvj7b6j/CB15-11-21-2019 v1.CSV?dl=0

 

[konjiki7]

@[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 8400-8430 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 : )