cancel
Showing results for 
Search instead for 
Did you mean: 

TUF B550m PLUS LLC controls

BuddyW
Level 7
I found the FAQ on the Digi+ VRM and that was very informative but it didn't cover much at all about the LLC. In particular, what does progressing from level 1 through 5 change; it's not even clear to me whether level 1 is 'more' or level 5 is 'more' or what 'more' even really means. And it's very hard to discern any change at all in the voltage (whether VCore or SVI2 TFN) since Ryzen CPU's are so darn dynamic and constantly dithering clock speed and voltage for individual cores even at idle, then lowering the voltage and clocks as heavier processing loads roll in.

Thanks for the info...
4,145 Views
7 REPLIES 7

RedSector73
Level 12
Level 3 is recommended by Yuri Bubliy (1usmus) for use with Clock Tuner for Ryzen (CTR) on ASUS boards, is the only thing ever come across about its use. Don't know how useful that is .... I read this to mean level 1 is lowest and five highest, but cant confirm this, just set level 3 as recommended.

Silent_Scone
Super Moderator
Hello,

The function of LLC is to scale the set voltage as close as possible whilst not allowing it to fall below the set voltage, counteracting vdroop and adjusting the amount of overshoot allowed. Level 1 is less load line calibration, level 5 is more.

Universally speaking when overclocking it is preferable to find stability with a higher voltage point than it is higher LLC to reduce both the amount and sustained overshoot.
13900KS / 8000 CAS36 / ROG APEX Z790 / ROG TUF RTX 4090

Silent Scone@ROG wrote:
Hello,
....
Universally speaking when overclocking it is preferable to find stability with a higher voltage point than it is higher LLC to reduce both the amount and sustained overshoot.


I am fully aware of the intended function of LLC. And as well, the effect of undershoot and overshoot (with roll-off and roll-on of heavy processing loads) on CPU stability. I just was not clear on how the scaling worked on the B550M TUF plus board I have. So it appears that level 5 introduces the maximum level of LLC compensation.

I'm also curios to know two other bits:

1: what exactly does the AUTO setting do? default to level 1? to level 5? use some logic to set a level based on VCore voltage setting?
2: is level 1 simply the lowest level of compensation? or might it be a negative level of compensation? This question is raised by 1Usmus' recommendation to use level 3, which might correspond to simply 'try to do nothing' or to let VDroop 'happen', so to speak. Bearing in mind that when left alone, the Ryzen boost algorithm pulls back voltage as the CPU heats up so maybe it's best to let it do it's own thing.

So, understand that I'm not interested in overclocking (pointless with Ryzen 5000, largely pointless with Ryzen 3000 and useless with my low-binned 3700X sample). But I've read several how-to's where they suggested just popping LLC to level 5 on their Asus boards simply to assure maximum performance even when not overclocking. Which, of course, seemed counterintuitive and that led me to question just what L5 actually meant.

BuddyW wrote:


So, understand that I'm not interested in overclocking (pointless with Ryzen 5000, largely pointless with Ryzen 3000 and useless with my low-binned 3700X sample). But I've read several how-to's where they suggested just popping LLC to level 5 on their Asus boards simply to assure maximum performance even when not overclocking. Which, of course, seemed counterintuitive and that led me to question just what L5 actually meant.




I'm not aware of what level is set by default, unfortunately. If you're aware of the function then knowing in which direction the loadline works is really all you should be concerned with.

There is even less point adjusting loadline than there is overclocking Zen if not wanting to do the latter. The 1usmus recommendation will likely be based on offering you the best possible chance of stabilising higher frequencies without exerting the limits of most individuals cooling solutions, and not be overly concerned with the negative impacts of removing vdroop.

You can monitor the behaviour of the CPU to a point including thermal changes, however much like your last thread one would need equipment with the adequate resolution and sampling rate for an accurate interpretation between various levels and VDROOP behaviour. Needless to say, on a stock system this is a huge waste of time.
13900KS / 8000 CAS36 / ROG APEX Z790 / ROG TUF RTX 4090

Silent
...howeve">Scone@ROG;834828]...howeve
r much like your last thread one would need equipment with the adequate resolution and sampling rate for an accurate interpretation between various levels and VDROOP behaviour...


As I said, I'm not interested in overclocking and definitely not extreme overclocking, such as would make above equipment an essential part of the process. I have watched a few of Buildzoid's vids (painful, but interesting and informative with respect to Ryzen's pecularities) and I do recognize how useful a decent 'scope could be. But all I am interested in is understanding just how my new motherboard functions...the better to tweak around with some things just to see how it affects it.

And by the way, being retired and house-bound by the pandemic time is one thing I've got. It is very much fun to delve into such esoterics so long as I don't blunder into a smoke releasing event.

BuddyW wrote:
As I said, I'm not interested in overclocking and definitely not extreme overclocking, such as would make above equipment an essential part of the process. I have watched a few of Buildzoid's vids (painful, but interesting and informative with respect to Ryzen's pecularities) and I do recognize how useful a decent 'scope could be. But all I am interested in is understanding just how my new motherboard functions...the better to tweak around with some things just to see how it affects it.

And by the way, being retired and house-bound by the pandemic time is one thing I've got. It is very much fun to delve into such esoterics so long as I don't blunder into a smoke releasing event.


Can appreciate the curiosity, but unless someone has done this groundwork already this behaviour is difficult to observe. Even by Buildzoids admission in some of his videos, the probes and scope being used aren’t necessarily adequate to observe certain transients when moving down the scale, so it’s pertinent to the question youre asking :).

These CPU are strung out right from the gate as you're aware and on a much smaller process than Intel counterparts. You certainly want vdroop here:o
13900KS / 8000 CAS36 / ROG APEX Z790 / ROG TUF RTX 4090

Silent Scone@ROG wrote:
.... You certainly want vdroop here:o


^^THAT^^ is one characteristic I have observed as it was quite obvious with the motherboard I was using before this.

The TUF seems to hold voltage a lot better, though, as it ranges from low-load/high boosting (where "up to 1.5V is to be expected") to extremely heavy all-core processing loads when seeing voltage around 1.3V (svi2 tfn) is really very nice. Holding a rock steady voltage is a good feature for conventional overclocking, but not since I want to run my 3700X in full-auto with some PBO that includes liberal use of scalar. I was kind of hoping there was a 'negative' LLC profile in that continuum.

EDIT: I seem to have found something that helps with this: CPU Current Telemetry Offset adjustment. In preliminary testing I'm dialing in a 10A positive offset, e.g., lie to the algorithm by telling it the CPU is drawing 10A more current than it actually is. It has very little effect at low loads as it's still boosting freely to max clocks and so won't hurt gaming. But it seems to be pulling back on voltage in heavy processing (Prime95), around 20mV or so. While doing that, it's holding core clocks at the same level. Probably because temperature wants to be lower with the lower voltage, I can only guess.

If this pans out with benchmarking and stability tests, it could be a hidden gem of an adjustment buried at the bottom of the AI Tweaking screen. Probably why the accountants at Asus missed it.

EDIT 2: it doesn't pan out. testing under heavy Prime95 brought voltage to the same average over a 5 minute span after 15 minutes of warm up.