Overclock.net banner
41 - 60 of 67 Posts

·
Registered
Joined
·
2,117 Posts
Discussion Starter · #41 ·
I meant limits are off so you're not bump into any. (mb) - means motherboard. Is there "motherboard" in a "PBO Limits" submenu on your board?
Per core temp sensors are in the latest HWInfo beta.
The idea was to obtain an average per core VIDs "calculated" by SMU without implicit restrictions applied.

As for prime, I ran OCCT with small/avx2 but overtemp protection I set was triggered at 90°+ that I would rather avoid tbh, its reeeeealy hot now :mad:
I may repeat it later when heat ceased a bit, or better on the next week (oracles promised < 30 at least)

View attachment 2515031
Re the Prime95 FFTs, as previously explained, its just to get a baseline of XXX voltage for XXX mhz.

What do you have vCORE set at in BIOS, 1.25v with droopy LLC ?

Whats interesting if the voltage is 1.25v is the temps you are getting into when running that test and yes I have understood that ambient temps are really high. I was hoping you would have run Prime95 so its a direct comparison rather than use OCCT (yes i know they both use the same algorithm for small/avx2).

For me 4600 is impossible with such little voltage, but even when using 1.31x temps are still not exceeding 90C with ambient temps in the 26C+

I will try with 1.25v OCCT to see max temps @ 4600 mhz ...

2515101
 

·
Registered
Joined
·
715 Posts
@mongoled
OCCT using prime95 as well
What do you have vCORE set at in BIOS, 1.25v with droopy LLC ?
Vcore set to 1.2375 "vdrooped" down to 1.206 with the strongest ("zero" load-line resistance) LLC available - that is ASRock boads VRM specifics, so there's no use of other than LLC 1 when statically overclocked. 1usmus mentioned that phenomena in his first ctr release notes as well.
I do have MSI board in office (actually typing this message on it )) ) and it's VRM behavior is nothing like ASRock one. Vcore stayed firmly as it was set even with 2nd strongest lvl.

even when using 1.31x temps are still not exceeding 90C with ambient temps in the 26C+
OCCT ran @ 33 ambient, that was quite extreme case for me, @ 26 i think temps wont cross 75.
26° - seems like frosts for Cyprus, ain't it? :)
 

·
Registered
B550 gaming edge, 5800x, 5600x, 3600xt, 1600AF, 2070S, 4x8 3200c14, 2tb adata 8200sx PRO, 500gb 970e
Joined
·
575 Posts
you guys with MSI boards, when you set 1.275V on the MAIN page,
for a manual OC did you check hwinfo and see if its set?
for whatever reason mines at 1.1V
so, i had to revert to the AMD overclocking section
and 4600 at 1.275V with 12 torture tests i fail. but at least now im not seeing 4600 at 1.1V
(svi2 was 1.1 and vid effective was 1.1 as well as per core VID was 1.1)

(EDIT) @mongoled
dont know why i can run my CO values how i do, but it would seem 4600mhz all the way up-to 1.35V i cant pass prime
insta fails almost. with auto LLC
so idk what to say at that point.
will check back when the 5800x comes in this weekend
as im off most the weekend and will hopefully get better luck out of the "core"
side and IMC as this 5600x has a good IMC but horrible CORE VF curves.
 

·
Registered
Joined
·
2,117 Posts
Discussion Starter · #44 ·
From discussion with @PJVol here

I re-ran the tests and the results are identical (within margin of error).

Re-flashed BIOS from Windows using the AFU utility, manually typed in all BIOS entries

BIOS: A.A4 (agesa 1.2.0.3B)
CPU/CPU NB LLC: Auto
Boost Override: 200 mhz
PBO: Motherboard Limits

CB RB23 run below

CO Disabled
Core0 (1st)Core1 (3rd)Core2 (5th)Core3 (4th)Core4 (2nd)Core5 (6th)
CO000000
VID1.2021.2291.2411.2551.2071.242

Photograph Light Table Rectangle Line

CO "Optimised"
Core0 (1st)Core1 (3rd)Core2 (5th)Core3 (4th)Core4 (2nd)Core5 (6th)
CO+5-3-7-6+6-9
VID1.2481.2351.2261.2471.2591.215

Photograph Light Table Rectangle Line

Now we do the same thing but for single core using the "p95_core_cycle" script

CO Disabled
Core0 (1st)Core1 (3rd)Core2 (5th)Core3 (4th)Core4 (2nd)Core5 (6th)
CO000000
VID1.3401.3111.3021.3011.3381.276
Frequency482047414712471348174649


CO "Optimised"
Core0 (1st)Core1 (3rd)Core2 (5th)Core3 (4th)Core4 (2nd)Core5 (6th)
CO+5-3-7-6+6-9
VID1.3261.3151.3161.3141.3241.321
Frequency476547644773476547504745


If any MSI users are able to do the same tests would be interesting to see if this is just an MSI thing or something that occurs with my setup...
 

·
Registered
Joined
·
2,117 Posts
Discussion Starter · #45 ·
you guys with MSI boards, when you set 1.275V on the MAIN page,
for a manual OC did you check hwinfo and see if its set?
for whatever reason mines at 1.1V
so, i had to revert to the AMD overclocking section
and 4600 at 1.275V with 12 torture tests i fail. but at least now im not seeing 4600 at 1.1V
(svi2 was 1.1 and vid effective was 1.1 as well as per core VID was 1.1)

(EDIT) @mongoled
dont know why i can run my CO values how i do, but it would seem 4600mhz all the way up-to 1.35V i cant pass prime
insta fails almost. with auto LLC
so idk what to say at that point.
will check back when the 5800x comes in this weekend
as im off most the weekend and will hopefully get better luck out of the "core"
side and IMC as this 5600x has a good IMC but horrible CORE VF curves.
No, never seen an issue with A.A4 BIOS when setting vCORE directly, has always set the correct value as per HWInfo64, however in early BIOS I have seen this issue before, I had to switch between AUTO, reboot, then try again and it worked...

My sample needs 1.31x volts for Prime95 Small FFTs, any higher and temps sky rocket fast!
 

·
Registered
Joined
·
715 Posts
CO "Optimised"
How did you get these ?
+5-3-7-6+6-9
I'd try +5 -1 -3 -6 +6 -3 (thats what I get, based on your data and using my approach)

do the same thing but for single core using the "p95_core_cycle" script
Is it just for science? Cause I don't see, how it may help you tune CO, because such workload distribution would hardly ever get encountered in real scenarios, and it doesn't take into account all-core interactions
 

·
Registered
Joined
·
267 Posts
The per core temp stuff is interesting. I don't have CB23 yet, but ran CB20 just to see. My coolest cores were 1 then 5, which are ranked 5th and 6th for performance, heh.
 

Attachments

·
Registered
Joined
·
2,117 Posts
Discussion Starter · #48 ·
How did you get these ?
+5-3-7-6+6-9
I'd try +5 -1 -3 -6 +6 -3 (thats what I get, based on your data and using my approach)


Is it just for science? Cause I don't see, how it may help you tune CO, because such workload distribution would hardly ever get encountered in real scenarios, and it doesn't take into account all-core interactions
No its not just for science

:D

Its for understanding how my cores act with regards to VID/Frequency and a specific workload AND to understand how there may be differences in the way VID is "distributed" depending on core loading scenarios, i.e. if its single or multicore.

Your point above is valid re "real scenarios" but I would still be blind to what's occurring if I didn't do the above tests !

Now to your tuned values, achievable all core frequency does not change, though VID values do (ive added difference in voltage in tables below)

CO "Optimised by PJVol"
Core0 (1st)Core1 (3rd)Core2 (5th)Core3 (4th)Core4 (2nd)Core5 (6th)
CO+5-1-3-6+6-3
VID1.2481.246 (+0.011v)1.248 (+0.022v)1.2471.2591.248 (+0.054v)

Photograph Light Table Rectangle Window

For single core using the "p95_core_cycle" script we see that there are some subtle differences and some that are quite profound, such as core5

CO "Optimised by PJVol"
Core0 (1st)Core1 (3rd)Core2 (5th)Core3 (4th)Core4 (2nd)Core5 (6th)
CO+5-1-3-6+6-3
VID1.323 (-0.003v)1.311 (-0.004v)1.310 (-0.006v)1.316 (-0.002v)1.326 (-0.002v)1.287 (-0.034v)
Frequency4758 (-6 mhz)4741 (-23 mhz)4756 (-17 mhz)4760 (-5 mhz)4754 (+4 mhz)4682 (-63 mhz)


My conclusion is as follows,

Because of the proximity of the cores to each other changes to the voltage in one core effects the characteristics of adjacent cores. I dont believe that this is down to the power budget but down to subtle changes in temperatures that seems to me to be the main driving force in "equalizing" the cores with regards to frequency/temp/voltage.

Because our CPUs, specifically talking about 5600x and 5900x, we will see very different characteristics in how the CPUs act due to the different core layouts, not to mention the different core characteristics.

Hence, tuning for CO needs to be done on a case by case basis.

Now back to the "is it for science" question, now that I know the baseline VIDs for my CPU in different scenarios, I can play with telemetry and see if this effects single core VIDs.

Its all about investigating what the CPU does in different scenarios, I now know that when I get my next Ryzen CPU that the first thing I will do is get my "baseline" figures so I can understand how the CPU acts.

:)

For the record I tuned my cores first by using core-cycler to see get peak frequency without crashing for each core independently, than used Y-Cruncher to fine tune the CO
 
  • Rep+
Reactions: PJVol

·
Registered
Joined
·
2,117 Posts
Discussion Starter · #49 ·
The per core temp stuff is interesting. I don't have CB23 yet, but ran CB20 just to see. My coolest cores were 1 then 5, which are ranked 5th and 6th for performance, heh.
AMDs "best cores" seem to be defined by which core can do X frequency with the least voltage, but this scenario does not dictate what we may see as the "best cores" with regards to most performance

:)
 

·
Registered
Joined
·
715 Posts
For the record I tuned my cores first by using core-cycler to see get peak frequency without crashing for each core independently, than used Y-Cruncher to fine tune the CO
You know, I have my doubts about the benefits of such testing (for the reasons given above), and wanna ask you in this regard, do you think those performant cores really need offset values that high (+5 or +6) ? How stable it will be in a real workload (without forced core/thread affinity) if you set CO values based on, lets say, 0 for the best cores, i.e. something like
0 -6 -9 -11 +1 -8 ?
 

·
Registered
Joined
·
2,117 Posts
Discussion Starter · #51 · (Edited)
You know, I have my doubts about the benefits of such testing (for the reasons given above), and wanna ask you in this regard, do you think those performant cores really need offset values that high (+5 or +6) ? How stable it will be in a real workload (without forced core/thread affinity) if you set CO values based on, lets say, 0 for the best cores, i.e. something like
0 -6 -9 -11 +1 -8 ?
I get WHEA 18 errors if the core0/4 have lower CO offset.

I couldnt work out a way (without using lower boost override) to stop this happening other than raising the CO which unfortunately knocks the frequency when running heavy threads on the cores.

But I may have to revist this as I always had "power supply current" set to AUTO, I am now setting this to "typical current" so may have to retest ...

** EDIT **
As ive been playing around with telemetry offset, I need to test to see if using the offset allows me to lower the CO on the good cores without having the WHEA 18 issues ....
 

·
Registered
Joined
·
715 Posts
As ive been playing around with telemetry offset, I need to test to see if using the offset allows me to lower the CO on the good cores without having the WHEA 18 issues ....
Or...
Can you try to set power limits to something like 115/70/300, and additionaly set cTDP to 95-100 and PPL то 115, and check, if that affected clocks, or score, whatever...

ps: In doing so, put emphasis on power reporting deviation preferably not to exceed 95-105% range
 

·
Registered
Joined
·
2,117 Posts
Discussion Starter · #53 ·
Or...
Can you try to set power limits to something like 115/70/300, and additionaly set cTDP to 95-100 and PPL то 115, and check, if that affected clocks, or score, whatever...

ps: In doing so, put emphasis on power reporting deviation preferably not to exceed 95-105% range
No difference

:/
 

·
Registered
Joined
·
715 Posts
Idk, having been testing cores while cycling heavy load, avx-like, noted 20+ W consumption per core - that is absolutely ridiculous and unrealistic for the Zen3
 

·
Registered
Joined
·
37 Posts
on my 5800X - when I increas EDC Limit to 300 i get much better AIDA L3 Cache Results. Checking HWInfo max EDC readings are 155A. Should i go with 300A set in BIOS or could I harm my CPU?
 

·
Registered
Joined
·
715 Posts
could I harm my CPU?
There's an internal edc limiter, so unless you're not in OC mode, it won't let your cpu consume potentially unsafe amount of power.
 

·
Registered
Joined
·
715 Posts
CPU enter OC mode when you set fixed allcore frequency and voltage. That basically disable CPB and most of PBO functionality, including limits (except Curve Optimizer, which should have been manually disabled).
 
  • Rep+
Reactions: sendap

·
Registered
Joined
·
8 Posts
on my 5800X - when I increas EDC Limit to 300 i get much better AIDA L3 Cache Results. Checking HWInfo max EDC readings are 155A. Should i go with 300A set in BIOS or could I harm my CPU?
Is there any other increase in Performance, or just in this AIDA Cache Benchmarkresult ?
 
41 - 60 of 67 Posts
Top