AMD Curve Optimizer any guides / experience

[jesdals]

My first couple of attemps with the curve optimizer in the PBO settings on the Aorus Master X570 ended with reboots and WHEAs

But did some more tinkering today and with PBO limits at 250 (W/A/A) and a per core negative on 5. I finaly got it stable - not sure what to think of the result and could use some guidance to go furter from here. Did try negative 10 on all cores first - but that did not fly well...

The core 0 result is up between my best, but still testing settings on my new AIO setup.

Did a GPUz bench - but nothing special there compared to earlier test

Had some better result in both single and multi results

Any one with some experience or knowledge about any good guides?

I tried raising the negative value of core 0 and 3 (the first and the fourth), but after 20 minutes it crashed and leaving core 0 on negative 10 was not possible - did give a higher boost and single core score ind GPUz

But unfortunately not stable

It did give a improved single core score

Now trying to increase the negative offset on the low boosting core did not seem to make any difference. The result below is with core 6, 8 and 10 at negative 10.

Changing the same low boosting cores to a positive 5 did not show much difference - but was not stable and crashed after 20-30 minutes

 

[jesdals]

Here is my latest PBO settings with F31 bios for Auros Master X570

 

[Zach_01]

Scalar X4 must push a kind of “stupid” amounts of voltage to the CPU cores. You should be aware and considered about this and the potential damage/degradation.
I wouldn’t use more than x2-3.

 

[jesdals]

Tuthfully theres not much guide to go by - based my settings on some of Builzoids rambling and test. But will try the 2x setting - first test with cpu benchmark in CPUid gives a bit lower cpu speed in multi and single core.

But you migth have a point about TDC and EDC limits - lowering them to 135 gave a supriceing change in mulit and single core performance

will do some stability test with these settings

Did a run with Hardware Info open, it gave above 4400 MHz all core and 4975 MHz single core - it does seem to improve results will do some more testing and try lowering the TDC/EDC even more
- apperently it alsow give lower cpu temps at max load

Result with Hardinfo running

It seems stable - lowering it to 130 and try stability here

130 was stable will try 120 now

Did gain some single core headroom

During stress test I saw these results cpu temp is interesting

Setting TDC/EDC to 100 raised single core performance but lowers multicore performance

will do some more testing of temps and stability

 

[juanyunis]

So I ended up with a pretty nice PBO setup with a 5950X and a X570 AORUS Master running latest F31. I have reached CPU-Z 680 single core and 13k multicore. Cinebench R20 11400 and Cinebench R23: 29100. My CPU never goes beyond 89C and also many cores are able to boost at 5-5.1Ghz, while the rest are able to go 4.9+. Using OCCT with Linpack test my all core clocks are between 4.4-4.5Ghz.

I live in a very hot and humid area, my ambient temperatures are about 28C at the moment but it can go as far as 34C, I'm not using any AC, just case airflow and 240mm AIO with push/pull configuration. I'm planning on switching to a 360mm AIO with push/pull too.

These are the settings:
CPU Voltage: Normal with +0.04378 Offset
CPU LLC: Auto
PPT: 270
TDC: 140
EDC: 140
Curve: All cores -25
Scalar: Auto
Max Boost: 50Mhz
Platform Temperature: Auto

 

[Zach_01]

While within operating temp of 95C it is still on the very high side at 89C.
@jesdals is on a much better spot with his last settings of 120A.
Look and compare his temps, Avg. and Max Core voltage (SVI2 TFN) and his readings on VRM, Power(POUT/Input).
Unfortunately his HWiNFO does not report any CPU PPT (W) or SMU(power) values but still through VRM (W) values you can directly compare it since you both have the same board.
VRM readings may be inaccurate but still comparisons can be made.
It would be better to be able to compare CPU PPT(W)+PowerReportingDeviation (when going 100% CPU load only) but still...

I dont know how exactly curve optimizer works since I lack a 5000 series CPU but take it easy with those voltages.
I would be pleased if anyone of you can show details and screenshots of the CurveOptimizer menu. Thanks!

 

[juanyunis]

@Zach_01 well, according to AMD technicians It is by design however if I can get the same performance with lower temperatures I will welcome that. Gonna try with TDC/EDC at 120.

@jesdals do you mind sharing your latest PBO settings like I did. Thanks.

 

[mtcn77]

PPT: 270
TDC: 140
EDC: 140

It is not how you do things. Also, 1.538v - I hope it is due to a fault in the sensor.
If you are looking for ST scores you clamp PPT, and otherwise for MT you keep PPT just 10% above EDC.
Those scores are too high. Better safe than sorry.

Ryzen 9 5950X Curve Optimizer to 5.1 GHz, PBO and overclocking - ElmorLabs

The much anticipated Zen 3 architecture has finally been launched close to a year and a half after Zen 2 while remaining on the same 7nm TSMC process. AMD's recently released Ryzen 3000XT showed there are frequency gains to be had just by manufacturing improvements over time. The question is...

www.elmorlabs.com

 

[Zach_01]

It is not how you do things. Also, 1.538v - I hope it is due to a fault in the sensor.
If you are looking for ST scores you clamp PPT, and otherwise for MT you keep PPT just 10% above EDC.
Those scores are too high. Better safe than sorry.

Ryzen 9 5950X Curve Optimizer to 5.1 GHz, PBO and overclocking - ElmorLabs

The much anticipated Zen 3 architecture has finally been launched close to a year and a half after Zen 2 while remaining on the same 7nm TSMC process. AMD's recently released Ryzen 3000XT showed there are frequency gains to be had just by manufacturing improvements over time. The question is...

www.elmorlabs.com

And actually his max Vcore is 1.556V as the SVI2 TFN is the most accurate sensor according to HWiNFO author. VID is just a request under a voltage/speed/load table.
And yet again what matters is the voltage under heavy and constant stress. This 1.55V could just be under very light load (low A) and short boost, as all Zen2/3 behave like this. Still it’s an indicator of higher voltages overall and I agree that it should be lower. Even with performance losses if necessary.

 

[jesdals]

Here is my settings - I do believe I will go up to 125 EDC/TDC

And my curve settings

 

[juanyunis]

I adjusted to 250/120/120 and I see good CPU-Z but cinebench scores dropped. While running OCCT I noticed that my all core clocks dropped to 4.3Ghz, temps are way better tho.

 

[jesdals]

Lowering the third setting to 130 did give better boost

Not still sure have to use the Elmo guide, but will test these settings some more - CPU-z result seems lower on multi about 4150MHz when running all core

It is not how you do things. Also, 1.538v - I hope it is due to a fault in the sensor.
If you are looking for ST scores you clamp PPT, and otherwise for MT you keep PPT just 10% above EDC.
Those scores are too high. Better safe than sorry.

Ryzen 9 5950X Curve Optimizer to 5.1 GHz, PBO and overclocking - ElmorLabs

The much anticipated Zen 3 architecture has finally been launched close to a year and a half after Zen 2 while remaining on the same 7nm TSMC process. AMD's recently released Ryzen 3000XT showed there are frequency gains to be had just by manufacturing improvements over time. The question is...

www.elmorlabs.com

Thanks for sharing that link to Elmo - havent found much about these settings

BTW when tinkering with the dual bios on the Gigabyte boards - do one save in the first slot as "bios 1" and if you end up in the other due to too optimistic setting make a "bios 2" saving there - then its more easy to find your way around I personally have a USB pen stored internally for bios and stuf - it makes thinks so much easier.

 

[Zach_01]

Here is my settings - I do believe I will go up to 125 EDC/TDC
View attachment 183248
And my curve settings
View attachment 183249

View attachment 183250

I adjusted to 250/120/120 and I see good CPU-Z but cinebench scores dropped. While running OCCT I noticed that my all core clocks dropped to 4.3Ghz, temps are way better tho.View attachment 183251

Thanks a lot for screenshots, I really appreciate it!
From what I see I think around 120-125A would be the best setting for performance, temperature and overall silicon stress.

 

[mtcn77]

I adjusted to 250/120/120

If you selected that, V²*I=(1.556V)²*120A=290.5watts, you are limiting your power just 16% at 250watts. You can do much better, in fact the reason 'the curve optimiser' exists is to help you do it.
You couldn't lower voltage bins quite as you could elevate them using the scalar option.
Cores don't need to have max boost, you cannot feed them the same power. Some are leaky, some are resistant. The gist is having the best of both worlds.

Thanks a lot for screenshots, I really appreciate it!

It really makes a difference that people follow through on such nuances.

 

[juanyunis]

After testing and testing all day I was able to find a sweet spot for my system.
Temps are "decent" never exceding above 80-81C while running cinebench r20/r23, cpu-z bench and OCCT (Linkpack and default). 9 of my cores were able to go between 5-5.1Ghz, the others were able to reach 4.9Ghz+.
CPU-Z scores and pretty good IMO, Cinebench R20 11200, R23 28700.

 

[Zach_01]

After testing and testing all day I was able to find a sweet spot for my system.
Temps are "decent" never exceding above 80-81C while running cinebench r20/r23, cpu-z bench and OCCT (Linkpack and default). 9 of my cores were able to go between 5-5.1Ghz, the others were able to reach 4.9Ghz+.
CPU-Z scores and pretty good IMO, Cinebench R20 11200, R23 28700.
View attachment 183277
View attachment 183278

Better...
Do you remember maybe what was the SVI2 TFN Core voltage during those 100% load benches/tests? I’m just curious!
There might be some difference between OCCT and CB R20/23 also.

 

[juanyunis]

These are my settings.

 

[Zach_01]

Thanks!
I was asking about HWiNFO reported “current” value of the “CPU Core Voltage (SVI2 TFN)” sensor at the time of the bench/stress test run.

 

[juanyunis]

@Zach_01 I just made a screen recording (The scores dropped I guess because i was running multiple apps and doing screen recording). Real time Cinebench R20 (Colors look washed out because my monitor is HDR)

 

[mtcn77]

@Zach_01 I just made a screen recording (The scores dropped I guess because i was running multiple apps and doing screen recording). Real time Cinebench R20 (Colors look washed out because my monitor is HDR)

Just do another test and do a screen capture at the right temperature that the actual test was running in order to find out about just how much it was tapping at the time. Power is related to the temperature.
Again, your results can get better if you lock the cores at a similar pace. There is no need for boost when running MT, forget it altogether really. You can get away with much lower EDC's than that. Remember, EDC is only necessary for single boost which out of 16 cores, it is a minor portion of PPT sum. If you allow your cpu to run cooler, you'll get the benefits even in ST loading.
Would you believe it when Intel was having the prime of its mobile division(before the axe fell) they were researching into melting thermal paste materials in order to boost for longer. It is all a condition of temperature management.

 

[Zach_01]

Just a screenshot after start and before it ends would be ok, no need to record.
Any way I saw it.

Roughly 1.1V, 125A and 169W PPT with a 105% PowerReportingDeviation.

The actual (true) power is 169/1.05= 161W

Applying voltage and current to @mtcn77 equation it comes close
V^2*I= (1.1*1.1) * 125 = 151W

Taking into account the accuracy and refresh/polling period of each sensor it adds up.

Thank you, that’s all I wanted.

 

[juanyunis]

Actually hw info locked up, but I can confirm that the voltage gets lower when doing multicore.

I need the single boost clocks for gaming. I use this computer for running some VMs 24/7 and heavy gaming. Some times for compiling code.

So what do you guys think I should do to have the best of both worlds multicore and single core?

So I ran another test and took 3 screenshots. Hopefully I did it right.

Screenshot 1 (After starting the test)

Screenshot 2 (In the middle of the test)

Screenshot 3 (At the end)

 

[bubbleawsome]

I'm new to Ryzen but I've got a 5900x I'm trying to tune some. I see EDC is the limiter for most situations, but would I want to keep it around 120A if I'm running the stock 1.5v maximum? It's closer to 1.22v under all-core load, but it will hit 1.48v under single core loads.

 

[juanyunis]

@bubbleawsome does your reach more than 1.5V? Or close like 1.498V? What is your motherboard? If you are running everything stock at the moment you can try enabling PBO in advanced mode and put some initial values, if your processor is 65W TDP then maybe you can try running PPT at 220, TDC/EDC at 120 and the curve -15, scalar auto and platform thermal auto, max boost at 0.

Edit: I though you had a 5800X, still those settings apply to you, you can increase the PPT to 240.

 

[Zach_01]

Actually hw info locked up, but I can confirm that the voltage gets lower when doing multicore.

I need the single boost clocks for gaming. I use this computer for running some VMs 24/7 and heavy gaming. Some times for compiling code.

So what do you guys think I should do to have the best of both worlds multicore and single core?

So I ran another test and took 3 screenshots. Hopefully I did it right.

Screenshot 1 (After starting the test)
View attachment 183301

Screenshot 2 (In the middle of the test)
View attachment 183302

Screenshot 3 (At the end)
View attachment 183303

The second is the one with max load, on the last was already finished.

I'm new to Ryzen but I've got a 5900x I'm trying to tune some. I see EDC is the limiter for most situations, but would I want to keep it around 120A if I'm running the stock 1.5v maximum? It's closer to 1.22v under all-core load, but it will hit 1.48v under single core loads.

Sounds about right... As for what is the best EDC value has no straight answer. It depends, mostly on temp.
From my understanding, using the curve optimizer with custom PBO limits you are by passing some of the protection of the chip. Mostly on current. It should be used with care and awareness of what is what. Temperature is an indication of stress but doesn’t tell the whole story. EDC too.

For example (and I’m going to use random numbers) having EDC, let’s say 140A at 70C is ok, but having 140A at 90C is not. The speed or voltage at those 2 cases is less significant but still play a role.
Same applies to every EDC value. It’s the combination of current(A), voltage and temp.

Keep temperature as far as possible from max operating limit (95C), under 80C I would suggest, EDC in reasonable levels (120~130) and all other (voltage, speed, watt) will fall were they should.

I’m not afraid of the 1.45-1.5V on single thread, nor the 200W PPT on multi.
It’s the high temp/EDC combination that can tear the chip’s internal traces rather quickly.

@bubbleawsome does your reach more than 1.5V? Or close like 1.498V? What is your motherboard? If you are running everything stock at the moment you can try enabling PBO in advanced mode and put some initial values, if your processor is 65W TDP then maybe you can try running PPT at 220, TDC/EDC at 120 and the curve -15, scalar auto and platform thermal auto, max boost at 0.

The 5900X has the exact same stock PBO limits with 5950X.

I believe those are

PPT: 142W
EDC: 140A
TDC: 90A