R5 5600 random frame rate spike in all games

[aska33j]

I understand what you’re saying and the logic behind it.
But I also understand that you do not own a Zen2/3 system. Otherwise you would know that max reported PRD value, like for example 150-700% does not occur under max CPU load, just because it’s max value. It doesn’t work that way.

Glad you found the solution.
It is said that SoC voltage on AM4 is safe up to 1.25V. Personally I don’t recommend anything above 1.2V.
At 1.15V you are perfectly fine.

If the issue is fixed now you can focus on tuning the system.
First I would test to see if what the BIOS is reporting as CPU power is true or not and how close is the reported value to reality.

1.
Use PPT:100W, TDC:70A, EDC:95A

2. Have HWiNFO opened and make visible the following sensors
a) CPU PPT
b) Power reporting deviation.

3.
Start Cinebench R23 or 24 on multithreaded mode

4.
A couple of seconds after CB starts click the clock icon on HWiNFO sensor window (bottom right) to reset all sensor values.

5. Wait a few minutes and before CB finishes, take the screenshot.

6.
Post the screenshot here or write what was the reported values of…
a) CPU PPT:___?
b) Power Reporting Deviation___?
…during the CB test

I would prefer a screenshot where the above with be visible during the test

here;

 

[Zach_01]

Looks like the CPU uses all the EDC (100%) it can and not reaching the power limit (PPT: 87-88%)
Plus the board BIOS slightly (~6%) overstates power (Power Reporting Deviation: ~106%)

This results the true power of the CPU to be
avg
PPT / PRD = ......
86.5W / 1.06 = 81.6W

New limits

PPT: 100W
TDC: 70A
EDC: 90A (down 5A from 95A)

...and see how this will change CPU behavior

 

[aska33j]

Looks like the CPU uses all the EDC (100%) it can and not reaching the power limit (PPT: 87-88%)
Plus the board BIOS slightly (~6%) overstates power (Power Reporting Deviation: ~106%)

This results the true power of the CPU to be
avg
PPT / PRD = ......
86.5W / 1.06 = 81.6W

New limits

PPT: 100W
TDC: 70A
EDC: 90A (down 5A from 95A)

...and see how this will change CPU behavior

here is the second test; And if this matters with the first limits you suggested ( 100/70/95 ) i got around 11000 multithread score and with this one i got 10646

 

[Zach_01]

You can play around with these settings and find what suits you in terms of performance and temperature.
Also you can use Curve Optimizer for potential better performance results.

The 5600 has a default boost limit of 4.5GHz and you already hitting that.
For more frequency you can use boost overdrive steps (+25~200MHz) but using this with negative curve optimizer can lead to instability.
No physical damage, just instability.

You can also try again to increase memory frequency but dont try at once the 3600MT/s

 

[aska33j]

You can play around with these settings and find what suits you in terms of performance and temperature.
Also you can use Curve Optimizer for potential better performance results.

The 5600 has a default boost limit of 4.5GHz and you already hitting that.
For more frequency you can use boost overdrive steps (+25~200MHz) but using this with negative curve optimizer can lead to instability.
No physical damage, just instability.

You can also try again to increase memory frequency but dont try at once the 3600MT/s

ill play around this these last power limits and see how it effects my gaming. i will leave the ram untouched for now lol.

as for the curve optimizer if i add +200mhz and put a negative offset this will lead to instability? and what if my 5 cores can go to -25 but one or two cores can only go -10 is this okay ?

And thank you so much for your help!

 

[Zach_01]

When using negative curve optimizer you're setting the CPU to work at the same speed with less voltage that the default, or work higher speed for the same voltage. Depends on the type of load (single thread, mid level, full load)
Adding more frequency (example +200MHz) to the curve can lead to not sufficient voltage for the corresponding frequency.
This is a "trial and error" process. There is no specific settings I can give you or predict now what the CPU can do.

Max negative curve optimizer steps are up to -30 and max frequency overdrive is +200MHz.
99.9% you can use both at max because there is too much clock stretching.

Remember that CurveOptimizer benefits single and mid-level loads like gaming and frequency boost overdrive benefits all core loads like cinebench MT or any other all core load.
You have to choose what suits you the most and maximize one of them, or pick the middle way for both.

As for how many negative steps each core can go depends on the CPU silicon. You have to find that by testing the limits of each core individually.
A good starting point is this:

See the individual "perf #n/n" (you have to have CPPC enabled in BIOS to see them)
First number (n/n) is the CPPC order of the cores
Second number (n/n) is the order Windows scheduler will choose to load after consulting the CPPC order.

Small numbers like 1, 2 are the cores that clock higher and usually loaded first (like on gaming) and higher numbered cores clock less.
Usually the higher clocked cores cannot take too much -CO steps and vise versa.

 

[aska33j]

When using negative curve optimizer you're setting the CPU to work at the same speed with less voltage that the default, or work higher speed for the same voltage. Depends on the type of load (single thread, mid level, full load)
Adding more frequency (example +200MHz) to the curve can lead to not sufficient voltage for the corresponding frequency.
This is a "trial and error" process. There is no specific settings I can give you or predict now what the CPU can do.

Max negative curve optimizer steps are up to -30 and max frequency overdrive is +200MHz.
99.9% you can use both at max because there is too much clock stretching.

Remember that CurveOptimizer benefits single and mid-level loads like gaming and frequency boost overdrive benefits all core loads like cinebench MT or any other all core load.
You have to choose what suits you the most and maximize one of them, or pick the middle way for both.

As for how many negative steps each core can go depends on the CPU silicon. You have to find that by testing the limits of each core individually.
A good starting point is this:

View attachment 368875

See the individual "perf #n/n" (you have to have CPPC enabled in BIOS to see them)
First number (n/n) is the CPPC order of the cores
Second number (n/n) is the order Windows scheduler will choose to load after consulting the CPPC order.

Small numbers like 1, 2 are the cores that clock higher and usually loaded first (like on gaming) and higher numbered cores clock less.
Usually the higher clocked cores cannot take too much -CO steps and vise versa.

hey ! how are you?

So ive been doing some testing with the CurveOptimizer and let me tell you its tedious and it really does take time.
after seeing many results online for cpus like mine and seeing what works and what doesnt for others. i dont think its worth it for my cpu and my usage to overclock and add the 200mhz boost.

With only pbo enabled 4.45ghz i get a single score in CPUZ of 615 and with 4.65ghz i get 645. Just more heat and more power consumption. with my mid-range 1080p pc im already getting more than 60fps in AAA games.

So for my testing;
Negative curve optimizer

for core 5 ( anything above -10 will result in error. at -20 the PC would shutdown ) i ran prime95 and this is what works for me. It took some time to know which core was failing since the pc would shutdown during test. No errors.

Also for the CPPC i dont think its working for me

CPPC ON. All perf Cores are 1

CPPC OFF.

So with CPPC Enabled is windows putting all work on 1 core or whats happening?

 

[Zach_01]

Don’t have a clue why all cores have #1 on the CPPC order. I think I never seen it before.

The concept of CPPC is that because not all cores can achieve the same frequency windows must know which are the best so on low/medium threaded tasks, including gaming, the CPU can achieve highest performance by loading those first.

In your case maybe the silicon quality is the same across all cores that’s why all are #1, or something else is going on. I can’t tell, as I said I never seen this before.

 

[natr0n]

turn off CO and disable SMT

see how that combo works for you.