Help me understand Ryzen 7900X3D behaviour/PBO

Hardware CPU
,
1

As some of you might remember, I’m not what you’d call an AMD fanboy. That being said, I’ve recently [3 days ago] finished a Ryzen 7900X3D build, and due to it being both a new system and the cache vs non-cache ccx’s, I don’t really know what “normal” is yet. Other than the fact I knew it was going to be a warm chip and it was OKAY that it was warm, I don’t know what would be considered correct in any other field [clocks, temps, which ccx is being used, etc], and I was all set to just not mess with anything when today something odd happened.

I have an EK 360 AIO in my build, which I knew again, wouldn’t do much save maybe a couple of degrees off the 89 TJMAX if the chip was being thrashed, but so far 82 degrees has been the peak so I’m happy. I also have an Aorus Elite AX in here, which has been working fine too.

although 82 is fine, when I heard about Eco Mode, I was interested to try it. But typically, I looked it up on Youtube first and found this article by OC3d that states putting Eco Mode on through Ryzen Master isn’t a good idea, and I didn’t fancy trying to do it manually, so that was that.

I did however go to the bios and have a look for Eco Mode beforehand, and couldn’t find it. I’m a very much leave it on Auto person, so I didn’t change anything, EXCEPT, as I’ve just discovered before writing this, I must’ve put PBO to Auto instead of off like I had it.

I noticed when playing a game tonight that the cpu temp was barely cresting the mid 60’s when before it would constantly be in the high 70’s, and my framerate was considerably lower in an area where it should’ve been much higher, so something wasn’t correct. I checked the bios, turned off PBO, and now it’s behaving as it had been before.

so although I originally wanted lower temps, I completely don’t understand what happened. Why did PBO on Auto DECREASE my temps? I didn’t successfully enable Eco Mode in the bios, so why did PBO I guess set it TO that? Isn’t PBO supposed to enable more performance???

the lower temps were nice, but what did I do? Why did it just drop by 15 degrees just by putting it to Auto? Basically I’m just confused why the temperatures I had accepted before have suddenly changed.

Edit: turns out I somehow installed 3.xx chipset drivers and not the 5.xx ones. I probably created this mess myself, and will see if it was just that mistake that was the anomaly.

1 Like
2

Only semi related but if you decide you don’t want to use curve optimizer and disable it, first set the number to 0, save, reboot then disable it was a bug for a while and idk if they fixed it yet, if you don’t the curve is kept and still applies

Point is, might be a bug about how the settings are saved

2 Likes
3

I’m not sure what OS your using, but I know Windows 11 needs the newest updates and the “xbox game bar” or what not to park and use cores properly. Maybe they fixed it in the drivers, but I noticed changes here. The BIO’s also makes a difference. Windows 11 in general I found utilizes cores better for even back to Ryzen 3000, my 3900XT was way cooler using Windows 11. I also get my AMD chipset drivers from AMD, not manufacturer as they can be a little behind sometimes. If I remember right my X670 chipset had an update late this or last month.

Some cpu bench/testing software wouldn’t utilize cores properly (IE for fastest cores should use second ccd with no cache) and game loads balance between the two. Also, the temp readings I have are very skewed… The GPU cores that are integrated in 7000 cpus (even when disbled) always were the hotest compared to CPU dies even with cache.

I have PBO at auto as well, Now I have an ASUS ROG STRIX X670E-E Gaming wifi which has a lot of built in optimizations and interesting features I need to dive into. (IE one feature is spart enough to know when to apply PBO for single thread loads vs an ALL core over clock for CPU intensive loads). Forget what it’s called.

My CPU maxes out at 75C. I was only able to break 85C with Prime 95 utilizing two cores right next to each other on one ccd (one without cache). Keep in mind I do have a custom loop with dual 360mm rads. So that may be a factor as well.

For me I did play with Curve optimizer. I had to do PER CORE. The die with vcache couldn’t take as much of a negative offset as the other ccd without. I ended up with -15 on the ccd with vcache, and -30 on the other. This could be optimized further but this is what it looked like on cinebench.

Baseline Stock:

Baseline CB
Baseline CB543×925 47.3 KB

-15/-30 Curve optimizer:

-30 CB
-30 CB473×1181 81.7 KB

A note on this!!! I was having issues where if the CO was too aggressive in the negative I would go to start cinebench on all core and PC would just reboot… NOW I did find a few corrupted windows files so I need to try again to see if that was a root problem. I just haven’t found time yet as I have been playing games when in front of my PC lately.

If these chips are anything like the previous generations they are so smart internally that a manual OC or messing with things will have a marginal return WITHOUT a lot of work and experimenting). The CPU already takes into consideration temp, voltage and speed to perform at it’s best… YET lowing power can have good results in better temps and speed depending on your silicon on your chip (Much like silicon lottery used to be…sad its gone :frowning: )

This is my own experience, I’m no engineer, but I can try things and learn from experimenting.

I was going to do a more detailed write up but I’m not sure if people would be interested or not. Either way happy to help if I can.

1 Like
4

I probably would enable a lot of this stuff if I had any idea I could do so safely and wasn’t worried about some rare wacky defaults that would do damage. Issue being I just don’t know anything about voltages or what IS safe enough to monitor that side of things.

1 Like
5

Okay so overclocking has gone completely backwards from the old days
Overvolting is not the move, undervolting is where it’s at so it’s impossible to fuck up your silicon

Positive curve is overvolting, I only see it useful if your chip has degraded
Negative curve is undervolting

The goal is to use less volts so you can use more amps in the same wattage envelope

Keep the heat in check and more amps won’t hurt

Higher clocks need more amps to drive them
They also have a voltage requirement that needs to be met so that’s why you can’t lower the voltage too much

There are 2 ways to overclock with PBO
Single core boost ore all core boost
For games you want single, for applications you want all core

For all core, you are power limited, so let’s reduce power consumption

  • Set PBO to advanced
  • Set PBO limits to mobo
  • Set a negative frequency to cap your single core boost, this doesn’t affect your all core boost for the most part since all core boost is so low in comparison
  • Set a negative curve starting with -10 working your way up, unlikely you’ll reach -30 but that’s the goal
  • Bench a variety of loads from all core to single, both avx and non

For single core, still want to make the core more efficient but we’re jacking up the frequency

  • Set PBO to advanced
  • Set PBO limits to mobo
  • Set frequency override to +200mhz
  • Set curve to -3 then work your way up
  • Testing non avx single/few core loads

Alternatively you can just reduce your power envelope instead of overclocking,

  • Set PBO to advanced
  • Set PBO limits to manual and desire power limits
  • Set frequency override to desired
  • Set curve to -5 and work your way up
  • Bench a variety of loads from all core to single, both avx and non
2 Likes
6

I appreciate that, man.

7

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.