5800x (and other Zen 3 chips) PBO settings/Temperature fix

[Mussels]

Not sure if its just margin of error but i get somewhat a little bit more score on R23 on my older bios (Beta January 2021) vs the newer one (september, MSI Mag B550M Mortar), or maybe perhaps i just missed to teak something, gonna take a look back on my other settings
PPT 120, TDC 85, EDC 110, CO Best core negative 5, 2nd best negative 8, the rest is 13, max temp 82C on a Corsair H60 push pull (ST: ~4850Mhz, MT ~4550Mhz - ~4650Mhz)
View attachment 223178

For reference, my last test yesterday got me 15,565 in R23

Hwinfo is so much better than CPU-Z, its not worth posting CPUz screenies these days

(Okay -30 DOES crash for me, but only at true idle. any kind of load is stable.)

 

[Athlonite]

For reference, my last test yesterday got me 15,565 in R23

that's a whole 100 points more than I currently get with a 4.55GHz all core OC

 

[Taraquin]

When I ran moneromining on my 5600X I set at tdp limit of 45W. Using 4000cl16 tuned ram, +200 pbo and -30x4 -29x2 on CO all core speed in CB23 was actually 3.7GHz, max temp is 53C. Didn`t expect that high frequency. Voltage during load stayed at 850-856mv. IO-die uses 20-21W so the cores actually run allcore 3.7 using 24-25W. That is impressive and very efficient! At same CO+PBO and stock 76W limit it runs at 4.55-4.6GHz allcore, 72C max temp.

 

[Ibizadr]

This is what I got with my CO and 10min of cinebench R23

 

[ViperXTR]

For reference, my last test yesterday got me 15,565 in R23

Hwinfo is so much better than CPU-Z, its not worth posting CPUz screenies these days

(Okay -30 DOES crash for me, but only at true idle. any kind of load is stable.)

Oh ok, i forgot i also have HWinfo installed on mine for a while now
Most apps remain stable on my 15-20 negative offset on curve optimizer, even cinebench, however some other CPU intensive apps such as RCPS3 PS3 emulator (while running god of war 3 in particular) are veery sensitive to CPU tuning, ive used this to determine stability of my system last time on tweaking curve optimizer and it's kind of reliable. RyujinX a switch emulator is same but not as cpu heavy.
I wish there are some sites who use emulators as well as benchmark tools, i remember anadtech used some dolphin emulation benchmark but that is already very old.
Some may argue about the stability of emulation but from what ive tested for years, they are pretty decent

 

[Oasis]

There has been a change of plans: post now, tune later. I wouldn't recommend too much perfectionism, it is a destination travelled not reached...

Very nearly done the memory testing around 1000% HCI Memtest, going for 1500% or so for my daily settings. should that be enough?
I will also post my Temps for my D15/NR200(P) with different PBO settings/CO/AutoOC

I got to 1460% at I got one error

Anyways onto PBO testing (125/90/125 w/AutoOC+200Mhz and CO -30 -23 -28 -11 -14 -24 -30 -21)
This is on a D15 and very low fan speeds (<1000RPM) (It's easily 5-10c less at max RPM)

 

[Mussels]

I may have dremeled out the mount on my EK block, used a derbauer offset mount and put liquid metal on...

PPT: 125W (seems stuck around 121W)
TDC:75A
EDC: 110A

15,739 in R23 at 66.5C

 

[ViperXTR]

I may have dremeled out the mount on my EK block, used a derbauer offset mount and put liquid metal on...

PPT: 125W (seems stuck around 121W)
TDC:75A
EDC: 110A

15,739 in R23 at 66.5C
View attachment 223359

So the offset mount really works nicely?

 

[Mussels]

So the offset mount really works nicely?

It helps a little, as does the LM

EK shrunk the thread for the screw holes, so it didnt fit without help




Needed some badly applied LM TIM:


And now she's ready to go

 

[ViperXTR]

It helps a little, as does the LM

EK shrunk the thread for the screw holes, so it didnt fit without help

View attachment 223360


Needed some badly applied LM TIM:
View attachment 223361View attachment 223362

And now she's ready to go

View attachment 223363

What's your temp before using this? Some Arctice Freezer users that also comes in with an offset mount only had few/minor degrees improvement

 

[Mussels]

What's your temp before using this? Some Arctice Freezer users that also comes in with an offset mount only had few/minor degrees improvement

It was on the previous page, around 77C

 

[Oasis]

I may have dremeled out the mount on my EK block, used a derbauer offset mount and put liquid metal on...

PPT: 125W (seems stuck around 121W)
TDC:75A
EDC: 110A

15,739 in R23 at 66.5C
View attachment 223359

Because Blender and R23 has basically the same temps as I found, isn't that a still a bit high for custom water cooling? As my little ITX PC with a D15 is only a few C higher while having higher power settings. I will test your PBO settings and R23 today

Also which chipset driver version are you on and fan speeds?

 

[Mussels]

Because Blender and R23 has basically the same temps as I found, isn't that a still a bit high for custom water cooling? As my little ITX PC with a D15 is only a few C higher while having higher power settings. I will test your PBO settings and R23 today

Also which chipset driver version are you on and fan speeds?

It shares with an RTX 3090, in aussie weather. 20C ambient is "cold" for the current season here. Fans and pump are static and dont change.

 

[mattferris]

With respect to curve optimizer, if it is the case that stability problems mostly occur at idle or near idle state, would setting a Windows power plan to High Performance instead of Balanced keep the processor out of one of the lower power states where instability occurs? I.e., would we be able to realize the efficiency of a higher curve offset at full boost without suffering the instability at idle?

 

[Mussels]

With respect to curve optimizer, if it is the case that stability problems mostly occur at idle or near idle state, would setting a Windows power plan to High Performance instead of Balanced keep the processor out of one of the lower power states where instability occurs? I.e., would we be able to realize the efficiency of a higher curve offset at full boost without suffering the instability at idle?

No, disabling C-states is a better choice there

 

[mattferris]

For reference, my last test yesterday got me 15,565 in R23

Hwinfo is so much better than CPU-Z, its not worth posting CPUz screenies these days

(Okay -30 DOES crash for me, but only at true idle. any kind of load is stable.)

Are you running similar/same power/current PBO limits to your original post, or something new that you'll share when you finish testing?

 

[Mussels]

Are you running similar/same power/current PBO limits to your original post, or something new that you'll share when you finish testing?

I've got that setting as my final OC, and i'm working on a power efficient one (although Gerkngs settings appear to be pretty good alread)

One efficient and one performance profile is all i'll need, since we have ECO mode for people who want to minimise it completely

 

[outpt]

I have one performance mode(4.95ghz) and using the motherboards 95 watt eco settings. Things are alright. I have no idea why my 5800x has decided to act right unless the psu was on the way out. Think I might play with my 5900x and see what it can do other than taking up desk space

 

[mattferris]

Hi. I have another question. If I set -30 curve offset on core 0, should the max Core 0 VID be lower than 1.500V by the offset? I'm trying to figure out how to tell if my curve offsets are actually taking effect.

 

[Mussels]

Hi. I have another question. If I set -30 curve offset on core 0, should the max Core 0 VID be lower than 1.500V by the offset? I'm trying to figure out how to tell if my curve offsets are actually taking effect.

View attachment 223765

I dont believe VID is altered, just voltage
So it may show 1.5V there, but voltage could be lower

I guess you'd notice by lower wattage consumed, or higher clocks achieved

 

[tabascosauz]

Hi. I have another question. If I set -30 curve offset on core 0, should the max Core 0 VID be lower than 1.500V by the offset? I'm trying to figure out how to tell if my curve offsets are actually taking effect.

View attachment 223765

Setting an offset can have an impact on VID, but VID doesn't mean anything on Ryzen. It's not what the cores are being fed.

You're also misunderstanding what CO is about. The CO offset shifts the V-F graph. What you're describing is the traditional Vcore offset, which would for example reduce your max Vcore to 1.47V for a -0.03 offset because it's applying the same, dumb, blind offset all the time, regardless of what the CPU is doing (so -0.03V at idle, same -0.03V at full load).

With a -30 offset you should easily be able to see a difference in benchmark scores. There should be little to no difference in observed power draw, that's not how CO works.

If you want to check voltage alone, go off of SVI2 TFN and load only one specific core (corecycler script, OCCT core cycler, etc). Watch the SVI2 TFN Vcore before and after to verify the offset is working.

 

[ViperXTR]

Is it me or after updating bios, my 5800X can't boost to 4.55 to 4.65Ghz all core anymore on C23, with same bios settings.

Edit: My bad, forgot to set the CO type from Positive to negative for individual cores. New Bios seems to make CO a bit more stable on higher negative values, RPCS3 in particular is very sensitive to it.

 

[mattferris]

I got adventurous with my system today. I have a 5800X seated on a CH8DH with a NH-D15 cooler. I ran a series of tests to try to hone in on a comfortable PBO setup. Here is what I found across four configurations. Temps were the max seen during a 10-minute CB23 MC run.

For each setup, curve optimizer was: -23 on best cores; -30 on "bad" cores
Initial Setup

• PPT: 142
• TDC: 95
• EDC: 125
• Max Temp Multi-Core: 82C
• CB23: 15134

Second Test

• PPT: 124
• TDC: 85
• EDC: 127
• Max Temp Multi-Core: 76C
• CB23: 14957

Third Test

• PPT: 120
• TDC: 75
• EDC: 110
• Max Temp Multi-Core: 69.8C
• CB23: 14653

Fourth Test

• PPT: 117
• TDC: 80
• EDC: 120
• Max Temp Multi-Core: 71.3C
• CB23: 14862

My board has something called dynamic OC switching where it can flip to a manual overclock when you cross a current threshold. I wanted to see how this stacks up against PBO so I came in initially with a 1.2V VID 46.5 multiplier (4.650GHz) switch. This seemed to work great in CB23. It was stable for a 30 minute run. Temps were in the 60s. It seemed like a slam-dunk.

Here's the issue. I loaded Prime95 torture test and temps went into the 80s and I had a hardware failure on core 0 (in Prime95, not a WHEA or anything Windows detected). Okay, so the system is not stable. I put the multiplier down to 44.5 and tried again. Prime95 is stable there, but there are two issues that I wanted to get your opinions on.
1. Temps were hitting 80C and were really unstable
2. I was watching Ryzen master during Prime 95 and TDC was 105A, which scares the hell out of me. How scared of this should I be?

PBO usually downclocks the chip massively during a prime95 run, but the manual OC just tries to do whatever you tell it to do. Should I just switch over to Test Four from above for the safety? The reason I tried out the auto OC switcher was because the voltages I was seeing from PBO were regularly in the 1.48V area (which sounds super high to me). But it looks like the trade-off is that PBO can keep you out of dangerous situations with loads that look like Prime95.

Sorry for the long post. I've been learning a ton lately and hope to pick a few more things up today.

 

[Mussels]

I got adventurous with my system today. I have a 5800X seated on a CH8DH with a NH-D15 cooler. I ran a series of tests to try to hone in on a comfortable PBO setup. Here is what I found across four configurations. Temps were the max seen during a 10-minute CB23 MC run.

For each setup, curve optimizer was: -23 on best cores; -30 on "bad" cores
Initial Setup

• PPT: 142
• TDC: 95
• EDC: 125
• Max Temp Multi-Core: 82C
• CB23: 15134

Second Test

• PPT: 124
• TDC: 85
• EDC: 127
• Max Temp Multi-Core: 76C
• CB23: 14957

Third Test

• PPT: 120
• TDC: 75
• EDC: 110
• Max Temp Multi-Core: 69.8C
• CB23: 14653

Fourth Test

• PPT: 117
• TDC: 80
• EDC: 120
• Max Temp Multi-Core: 71.3C
• CB23: 14862

My board has something called dynamic OC switching where it can flip to a manual overclock when you cross a current threshold. I wanted to see how this stacks up against PBO so I came in initially with a 1.2V VID 46.5 multiplier (4.650GHz) switch. This seemed to work great in CB23. It was stable for a 30 minute run. Temps were in the 60s. It seemed like a slam-dunk.

Here's the issue. I loaded Prime95 torture test and temps went into the 80s and I had a hardware failure on core 0 (in Prime95, not a WHEA or anything Windows detected). Okay, so the system is not stable. I put the multiplier down to 44.5 and tried again. Prime95 is stable there, but there are two issues that I wanted to get your opinions on.
1. Temps were hitting 80C and were really unstable
2. I was watching Ryzen master during Prime 95 and TDC was 105A, which scares the hell out of me. How scared of this should I be?

PBO usually downclocks the chip massively during a prime95 run, but the manual OC just tries to do whatever you tell it to do. Should I just switch over to Test Four from above for the safety? The reason I tried out the auto OC switcher was because the voltages I was seeing from PBO were regularly in the 1.48V area (which sounds super high to me). But it looks like the trade-off is that PBO can keep you out of dangerous situations with loads that look like Prime95.

Sorry for the long post. I've been learning a ton lately and hope to pick a few more things up today.

Added to the front page, under a spoiler in the 5800x section

I dont think we really know what the limit for amps are with danger on these chips, tbh

Tested out what dropping the values 10% would do:
over 10C drop

Combining the remount with LM and that 10% drop in values, i'm down to 65C full load in R23, and lots lower in everything else

 

[mattferris]

I have an update on mine.

I have to confess, first, that I didn't believe Mussels on not using the curve optimizer in Extreme Tweaker. All of my other PBO settings were working fine from the Extreme Tweaker menu, so I assumed my Curve Optimizer settings were in effect. Out of curiousity, I moved all of my settings over to the AMD Overclocking menu. My voltages are much lower and I'm boosting 200MHz higher than I was before.

On the settings from my fourth test (above), my CB23 score went from 14862 to 15541.

I can't stress this enough. If you are using a Crosshair VIII Dark Hero, avoid the PBO menu in Extreme Tweaker.

Edit: And this means I probably need to stress test to see if my previous curve offsets are stable, since they were never working in the first place.