Overclock.net banner
1 - 15 of 15 Posts

·
Registered
5950X
Joined
·
47 Posts
Discussion Starter · #1 · (Edited)
Hi!

I recently came to near end with memory OC - 3733MHz 14-14-14-28, subtiming optimized (IMC won't let 3800MHz+ to be stable or with no WHEAs). Here are my timings and full story if you are interested:
2519265

My full PC specs are:
CPU: Ryzen 5950X @ stock
CPU cooler: Kraken X73 @ top exhaust, Thermal Grizzly Kryonaut Extreme paste
GPU: RTX 3090 TUF OC @ MSI Afterburner auto OC custom curve, 1920MHz core max, +200MHz memory, +7% power limit, V3 BIOS
RAM: 2x16GB 3733MHz CL14-14-14-28, subtimings tweaked @ 1.45V (G.SKILL F4-4000C14D-32GTZN, 4000MHz CL14 1.55V)
Mobo: X570 Aorus Master v1.2 @ F34 BIOS
SDD (main): WD SN850 2TB, PCIe 4.0
SSD (storage): RAID 0 -> 2x XPG SX8200 Pro 2TB, SM2262ENG controller, PCIe 3.0
PSU: BeQuiet Dark Power Pro 12 1200W
Other: Sound Blaster AE-9
Case: Phanteks P500A with 3 front intake, 1 rear exhaust 140mm fans
OS: Windows 10 Pro 64-bit

As for now I opt for OCing my Ryzen 5950X - and I mostly aim for trying undervolting by curve optimized.

I read some about how to do it, how to check which core failed etc., but I still have some questions bugging me:


1) Could undervolting with CO shorten my CPU life?

2) Could undervolting with CO degrade my CPU performance?

3) Could undervolting with CO make IMC less stable (so my FLCK won't be able run with CO FCLK at 1866MHz)?

4) Should I start with searching for best manual "PBO limits" or with CO undervolting?

5) Does it make sense to do CO undervolting on stock settings (no PBO or manual "PBO limits")?
 

·
H₂O Aficionado
Joined
·
4,763 Posts
What causes any CPU to shorten its life or performance? If you ask that question, you'll quickly find that its voltage, heat, and frequency (usage in the context of CPU life).

One of the only risks I am aware of with undervolting (via any method) is instability. If your CPU isn't supplied enough voltage for the request of the task, the CPU will crash due to error.

With the way Zen works, you're typically constrained due to power and heat.
Undervolting is a safe way to help improve performance. Here are your options in a nutshell:
  1. Maintaining or increasing voltage and increasing amperage (TDC/EDC) and wattage (PPT) limits may increase clock speeds but you will face or impose thermal limits with conventional air or water cooling.
    1. Best approach if thermals are not a constraint or concern.
  2. Lowering voltage but maintaining PPT/TDC/EDC will result in better temperatures, which may bring up the frequency.
    1. Best approach if thermals are your biggest constraint or concern.
  3. Raising your PPT/TDC/EDC limits (PBO options) and lower voltage via curve optimizer.
    1. Best balanced approach.
That should help answer 1 & 2.

For 3, 4, and 5, I will let others chime in.
 

·
Iconoclast
Joined
·
31,435 Posts
Raising your PPT/TDC/EDC limits (PBO options) and lower voltage via curve optimizer.
Curve optimizer, in and of itself doesn't alter voltage, it alters the boost frequencies at a given voltage. If frequency is capped via something else, then yes, voltage will also go down, but if allowed, voltage will scale normally and boost speeds will increase with the larger negative offsets.

Even on air, this is what happens with voltage on auto and negative offsets with my 5800X at +200MHz. It will always top out at a peak VID of ~1.5v, but it will reach max boost more often on single cores and hold a higher boost clock on all core loads.
 
  • Rep+
Reactions: Slaughtahouse

·
H₂O Aficionado
Joined
·
4,763 Posts
Thanks for the clarification :) I thought the adjustments, per core or per entire curve, were mV adjustments.

E.g. If you applied -10 in core optimizer, you are adjusting the curve by negative 10mV. So instead of one core requesting 1.5v for 5.0 GHz, it would now request 1.49v.

I don’t have Zen 3 (3800X owner) but that was my interpretation based on my readings.
 

·
Iconoclast
Joined
·
31,435 Posts
Thanks for the clarification :) I thought the adjustments, per core or per entire curve, were mV adjustments.

E.g. If you applied -10 in core optimizer, you are adjusting the curve by negative 10mV. So instead of one core requesting 1.5v for 5.0 GHz, it would now request 1.49v.

I don’t have Zen 3 (3800X owner) but that was my interpretation based on my readings.
It's more complex than that. The offset shifts the whole curve for that core, but it doesn't directly translate into a voltage change unless you happen to be running a fixed clock speed. You can even have a negative offset increase voltage if you increase the boost clock override at the same time, or reduce CPU temperature. The only sure thing is that voltage for a given clock speed is lower.


Behavior is still similar to what Elmor describes there, though more recent AGESA versions seem better at adhering to the 1.5v cap in the absence of direct voltage manipulation (i.e. not the curve, but an actual core voltage setting), at least on the boards I've used with 5000 series parts (ASRock, MSI, and Gigabyte).
 

·
H₂O Aficionado
Joined
·
4,763 Posts
Thank you for the additional context and explanation.
 

·
Old to Overclock.net
Joined
·
361 Posts
Wanna add my 5c to this topic - I wonder how Vcore offset works with CO,
and is it possible to use both dynamic OC feature and CO?
 

·
Registered
Joined
·
387 Posts
My experience with pbo curve has been a bit interesting. I don't know why, but I could never get it to work right, and I don't know if it's because I cap my cpu voltage to 1.27... Or maybe an issue with my board or bios....

Either way I was able to create some nice performing and stable settings.... But after roughly two days time my PC would start crashing back to back to back until I cleared cmos. Even more conservative settings with pbo curve yielded the same results... So it really seemed if i touched it at all, it woukd create issues. But would run perfect for ~2 days? I don't get it.
 

·
Overclock the World
Joined
·
3,126 Posts
Wanna add my 5c to this topic - I wonder how Vcore offset works with CO,
and is it possible to use both dynamic OC feature and CO?
Yes
You can go -30 and +85mV positive vcore offset
But performance won't be thaat great
The best is to tune CO with CTR with OB feature disabled.
Here is a cheat sheet
2519940

here is tool.exe
Bios Modding & OC - Google Drive and other fun stuff

Keep in mind that unlocking EDC, does allow Cache to fully overdrive till FIT limits it
But this also allows more constant current to pass - soo perf is lower
This one you have to limit back with lower TDC
Here some more reading about "limits"
Post CoreCycler - tool for testing Curve Optimizer settings & CoreCycler - tool for testing Curve Optimizer settings this :)

ASUS DynamicOC trigger, does run OC mode shortly, which is a good way to bypass EDC FUSE Limit
But generally limits do apply and CO is a magnitude changer, not an overrider
Project Hydra (1usmus) will show it more clearly , once it's out away from Patreon
 

·
Old to Overclock.net
Joined
·
361 Posts
Yes
You can go -30 and +85mV positive vcore offset
But performance won't be thaat great
The best is to tune CO with CTR with OB feature disabled.
Here is a cheat sheet
View attachment 2519940
here is tool.exe
Bios Modding & OC - Google Drive and other fun stuff

Keep in mind that unlocking EDC, does allow Cache to fully overdrive till FIT limits it
But this also allows more constant current to pass - soo perf is lower
This one you have to limit back with lower TDC
Here some more reading about "limits"
Post CoreCycler - tool for testing Curve Optimizer settings & CoreCycler - tool for testing Curve Optimizer settings this :)

ASUS DynamicOC trigger, does run OC mode shortly, which is a good way to bypass EDC FUSE Limit
But generally limits do apply and CO is a magnitude changer, not an overrider
Project Hydra (1usmus) will show it more clearly , once it's out away from Patreon
Thanks, now only need time to dig in it :D

BTW, patch C seems like solved the problem 2nd best core always boosting higher.

But it is still confusing, why then all cores hit max boost if AMD's cpus have cores "binned" from fab?
2519985
 

·
Registered
Joined
·
714 Posts
why then all cores hit max boost if AMD's cpus have cores "binned" from fab?
Chips are binned, not cores.
Run OCCT large/avx/extreme/steady in per-core cycling mode and keep track of the frequency each core is run at.
 

·
Registered
Joined
·
714 Posts
PBO doesn't work on static/capped vcore voltage..
Aside from absurdity of the original post, it raises an interesting question, though.
PBO doesn't work with static OC. And it seems logical to assume that Curve Optimizer also follows this rule, but does it really?
 

·
Registered
Joined
·
387 Posts
PBO doesn't work on static/capped vcore voltage..
Do you mean cpu core voltage or vcore soc? I kind of figured static core voltage might mess with it, but wouldn't expect vcore soc to
 

·
Cheesebumps!!
Joined
·
2,783 Posts
PBO doesn't work with static OC. And it seems logical to assume that Curve Optimizer also follows this rule, but does it really?
on newer AGESA's yes this won't work on older ones it somehow works along with CO upto the point where you capped the voltage, also with the exception of the ROG Dark Hero which combines both static OC and PBO features.

newer firmware's really nerfed the Day 1 performance of the Ryzen 5000 series..if I revert to the earlier BIOS from last november 2020 I would almost see my frequencies go high up as intended with the PBO settings..on newer ones, it just sticks to the "plan" most of the time and the intended boost override you made is merely being seen if you monitor em..

Do you mean cpu core voltage or vcore soc? I kind of figured static core voltage might mess with it, but wouldn't expect vcore soc to
static Vcore..VSOC neither has any bearing as its for the IMC
 
1 - 15 of 15 Posts
Top