7800X3D curve optimizer per core

[smoothnobody]

i stress tested every core, one by one, and found the sweet spot for every core. but when i run an all core stress test, it crashes instantly. i don't understand why.

[Dedayog]

1 hour ago, smoothnobody said:

i stress tested every core, one by one, and found the sweet spot for every core. but when i run an all core stress test, it crashes instantly. i don't understand why.

What are the per core values vs the all core value?

[smoothnobody]

30
29
21
24
16
22
28
29

 

all core 19

[Dedayog]

1 minute ago, smoothnobody said:

30
29
21
24
16
22
28
29

 

all core 19

Your all core exceeds one of your single cores, most likely causing your issue.

 

Can you run all core at -15?

[smoothnobody]

i must not be understanding something, or maybe i did not provide enough information and we are not on the same page. i am not trying to push the frequency higher. i am undervolting. my multiplier is auto. my curve optimizer is negative value. bigger number means less voltage right? if i can run -19, why would i want to run -15?

 

if my understanding is on point, i was confused why that one core when stressed by itself wants -16, but when i stress all the cores it's happy at -19. 

[Dedayog]

9 minutes ago, smoothnobody said:

i must not be understanding something, or maybe i did not provide enough information and we are not on the same page. i am not trying to push the frequency higher. i am undervolting. my multiplier is auto. my curve optimizer is negative value. bigger number means less voltage right? if i can run -19, why would i want to run -15?

Because one of yours cores can't run -19, it can only go -16.  You're trying to use less voltage, but that core needs more than that.

 

If voltage is 100 base, you're trying to push only 81 (with a -19).  But that core says it can do -16, which is 84 volts.   You're starving that core and it may be causing the issues.

 

Don't forget, each core needs a certain amount of power to run properly.  You can give too little or too much, and cause problems.  

 

[smoothnobody]

what do you mean one of my cores can't run -19? i'm not using -19 on any core when using per core. that -16 per core runs fine at -19 when all core. i don't understand why that one core runs fine with less voltage when all core but wants more voltage when per core. 

 

figured it might be helpful to share my undervolt method. when adjusting per core, i start with all the cores at -0. i then stress 1 core at a time, starting at -30, working my way down -29, -28 -27 etc. once it passes 1 hour stress test, i move on to the next core. when i get through every core, i repeat 1 core for 1 hour stress test with several passes. when i stop seeing errors, i switched to core cycling for more hours. is this the right way to do this?

[-rascal-]

 

What?

Your statement contracts your starter post.

 

21 hours ago, smoothnobody said:

i stress tested every core, one by one, and found the sweet spot for every core. but when i run an all core stress test, it crashes instantly. i don't understand why.

When you run all core -19, it instantly crashes.

 

18 hours ago, smoothnobody said:

i must not be understanding something, or maybe i did not provide enough information and we are not on the same page. i am not trying to push the frequency higher. i am undervolting. my multiplier is auto. my curve optimizer is negative value. bigger number means less voltage right? if i can run -19, why would i want to run -15?

 

if my understanding is on point, i was confused why that one core when stressed by itself wants -16, but when i stress all the cores it's happy at -19. 

18 hours ago, smoothnobody said:

what do you mean one of my cores can't run -19? i'm not using -19 on any core when per core. that -16 per core runs fine at -19 when all core. why does it run with less voltage when all core but wants more voltage when per core? both loads are 100% on that core.

 

not sure if relevant, but since this is my first time playing with curve optimizer, i figured it might be helpful to share my method. when adjusting per core, i start with all the cores at -0. i then stress 1 core at a time, starting at -30, working my way down -29, -28 -27 etc. once it passes 1 hour stress test, i move on to the next core. when i get through every core, i repeat 1 core for 1 hour stress test with several passes. when i stopped seeing errors, i switched to core cycling for more hours. is this the right way to do this?

But now you are saying -19 IS okay. 

 

The lower the number (e.g. -19 vs -16) the lower the voltage, yes.

You need to also understand, putting ALL cores under stress/load will have a different overall effect on the CPU compared to only ONE core under load/stress. When ALL cores are under stress, you are requesting more current, more of the CPU (e.g. cache, Infinity Fabric) is used, more voltage droop, etc.

[smoothnobody]

-19 all core stress test is stable.

 

with this curve, per core stress test is stable, all core stress test is not stable.

 

30
29
21
24
16
22
28
29

[Dedayog]

7 hours ago, smoothnobody said:

-19 all core stress test is stable.

 

with this curve, per core stress test is stable, all core stress test is not stable.

 

30
29
21
24
16
22
28
29

How is all core stable, and not stable, at the same time?  What all core value isn't stable?

[smoothnobody]

all core stress test is stable when all the cores are at -19........

 

core 0  -19
core 1  -19
core 2  -19
core 3  -19
core 4  -19
core 5  -19
core 6  -19
core 7  -19

 

all core stress test is not stable when setting each core individually.......

 

core 0  -30
core 1  -29
core 2  -21
core 3  -24
core 4  -16
core 5  -22
core 6  -28
core 7  -29