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

[Taraquin]

That would be great only I have no Idea what to look for or where to look, or how to understand what I click on when I do find something, most of it looks like binary gibberish and I have no idea how to read binary.

You must find the windows logs - system. Find the time of the reboot and you can see what core crashed. Generally the report as WHEA18. If I were you I would start with a baseline like -10 on all. Find which ones crash, the ones that don`t try running -15 or higher. Corecycler can be useful for many

 

[tabascosauz]

That would be great only I have no Idea what to look for or where to look, or how to understand what I click on when I do find something, most of it looks like binary gibberish and I have no idea how to read binary.

Event Viewer > Custom Views > Administrative Events > sort by Source column to find WHEA-Logger events

For core-related errors you're looking for either Unknown Source or Cache Hierarchy as the Error Type. The APIC ID will tell you what core is the problem - iirc it doesn't count SMT threads (don't quote me on that) and starts from 0 like most programs, so you'd be looking at a number between 0 and 5.

The one below is a Bus/Interconnect so a different IF-related type, and not core-specific so APIC ID 0

 

[oobymach]

I have no event id 18 nor whea logger events in the requested folder, there are 39k events in the list, my whea list is error free.

 

[Taraquin]

I have no event id 18 nor whea logger events in the requested folder, there are 39k events in the list, my whea list is error free.

View attachment 239423
View attachment 239424

Try to find what kind of error you get when the bsod occured

 

[freeagent]

Ppt is important for getting single core up, as well as for sustained all core loads. In my limited experience curve is based off of your ppt tdc and edc settings. If those aren’t right then curve won’t work well.

 

[outpt]

I have no event id 18 nor whea logger events in the requested folder, there are 39k events in the list, my whea list is error free.

View attachment 239423
View attachment 239424

Look under txt in cpu-z it will tell you what core crashed.

 

[freeagent]

In my limited experience curve is based off of your ppt tdc and edc settings. If those aren’t right then curve won’t work well.

My curve isn't perfect, but its fairly tight overall I think.



Edit:

Playing with Core Cycler now..

Edit again:

Well, it’s just as I suspected. My CPU is shit. It won’t even do +50 with an all core -15 or 10.. which loads it at 4950 in core cycler. But for everything else my original settings were cock solid, except core cycler. I could bench, fold, crunch something.. game do whatever with no errors, bluescreams, hardware errors.. nothing

The last time I played with CC I think it failed at stock settings.

Should I RMA it?

 

[Mussels]

seems like the curve just gives a more effect clock speed on all cores doesn't seem to do much for single thread for me.

Yes, that's what it does.

It lowers the wattage/amperage at any given clock speed, and those limits are most easily reached multi threaded.


Free: personally, i don't believe in RMA'ing because it wont overclock.
-15 is a large undervolt, few chips are stable past -10 (remember they're unstable at IDLE, so many people make bullshit claims and say its stable when its not)

 

[freeagent]

Yes, that's what it does.

It lowers the wattage/amperage at any given clock speed, and those limits are most easily reached multi threaded.


Free: personally, i don't believe in RMA'ing because it wont overclock.
-15 is a large undervolt, few chips are stable past -10 (remember they're unstable at IDLE, so many people make bullshit claims and say its stable when its not)

Ok good, thank you

I wont RMA it because it was working fine for the way I used it. I have flashed to the latest AGESA, went back to a clean install of 11. Boost and voltage are waaay down, but it still runs ok. I am not going to fuss over it.. I ran it completely stock and it felt just fine. 1600 1:1, factory limits and boost.. it was ok. Games were fine, it wasn't a massive loss of performance.. because I play at 60fps I didn't notice any loss lol.. Power has gotten a lot more expensive here, and will continue to do so..

 

[oobymach]

Thanks for trying to help, now testing with -10 -5 -5 -10 -15 -10 after crashing more aggressive undervolts on the curve all over, I'm learning this new shit as I assume many of you already did by trial and error, figuring out what I can do and what I can't by testing both low and heavy loads and figuring out what works and what doesn't.

Passed 12 rounds of p95 because I forgot it was on.

 

[DemonicRyzen666]

Thanks for trying to help, now testing with -10 -5 -5 -10 -15 -10 after crashing more aggressive undervolts on the curve all over, I'm learning this new shit as I assume many of you already did by trial and error, figuring out what I can do and what I can't by testing both low and heavy loads and figuring out what works and what doesn't.

Passed 12 rounds of p95 because I forgot it was on.

like mine error on core 3 with in 3 minutes in prime95
Later on, error on core 6 with 18 minutes in prime95.
These chips are so weird.

 

[Athlonite]

My curve isn't perfect, but its fairly tight overall I think.

View attachment 239462

Edit:

Playing with Core Cycler now..

Edit again:

Well, it’s just as I suspected. My CPU is shit. It won’t even do +50 with an all core -15 or 10.. which loads it at 4950 in core cycler. But for everything else my original settings were cock solid, except core cycler. I could bench, fold, crunch something.. game do whatever with no errors, bluescreams, hardware errors.. nothing

The last time I played with CC I think it failed at stock settings.

Should I RMA it?

No just stop using Core Cycler as obviously that's the issue

 

[freeagent]

No just stop using Core Cycler as obviously that's the issue

I did, but again like the last time it kind of bugs me.

I also went back to the old voltage pumping bios.. still on 11 though.. so pretty.

 

[Mussels]

I'm thinking the lowered voltage/amperate is related to the intended changes for the 5800x3D

 

[ntdouglas]

+200
-15 crashes any higher
-19 fastest core
-23
-19 fastest core
-23
-23
power limits
130
80
130
these power settings seem high compared to others but running occt the ppt and the 3rd one (edc?) are close to maxing out. Single core boost isnt quite what it was before this journey but temps are down 10-12 degrees. Multi core is pegged at 4.74 at 72c though. One run of occt I know I seen cores 2 and 4 flash 4.95 for a second or two. I want to get to 5ghz single core. Any ideas

 

[DemonicRyzen666]

+200
-15 crashes any higher
-19 fastest core
-23
-19 fastest core
-23
-23
power limits
130
80
130
these power settings seem high compared to others but running occt the ppt and the 3rd one (edc?) are close to maxing out. Single core boost isnt quite what it was before this journey but temps are down 10-12 degrees. Multi core is pegged at 4.74 at 72c though. One run of occt I know I seen cores 2 and 4 flash 4.95 for a second or two. I want to get to 5ghz single core. Any ideas

The only way to get higher on a 5600x is to use a slight bump in bus speed. Which means you have to lower you ram divider before you even try a higher bus speed to counter act it. It won't take much to get a decent bump anyways even something like 100.03 is still a bump for Ryzen. I know I got pretty high on my bus speed, but I'm not using an NVMe SDD. I'm using a Sata SSD. I'm not sure what the usually is for high bus speed, but you don't need to get anywhere near 101.0 on bus speed. Tiny bus speed increments in Ryzen can throw up a lot of speed increase.

 

[ntdouglas]

Would raising it slightly effect nvme drives?

The only way to get higher on a 5600x is to use a slight bump in bus speed. Which means you have to lower you ram divider before you even try a higher bus speed to counter act it. It won't take much to get a decent bump anyways even something like 100.03 is still a bump for Ryzen. I know I got pretty high on my bus speed, but I'm not using an NVMe SDD. I'm using a Sata SSD. I'm not sure what the usually is for high bus speed, but you don't need to get anywhere near 101.0 on bus speed. Tiny bus speed increments in Ryzen can throw up a lot of speed increase.

Even with pbo limits disabled?

 

[freeagent]

I dropped my 5600X in and ran CC with no problems, I guess my 5900X just does not like that program..

Core 3 was cranky with 4850, so I chose +150 instead..

 

[oobymach]

I dropped my 5600X in and ran CC with no problems, I guess my 5900X just does not like that program..

Core 3 was cranky with 4850, so I chose +150 instead..

View attachment 239706

I tested with CC and had a core 2 fail with +200 even when undervolt was 0 for that core (-5 0 -5 -10 -15 -10 curve) so I'm just going to ignore it since 12 rounds of regular p95 passed. CC is large fft, my cpu ran at 71 degrees during the test.

 

[freeagent]

I tested with CC and had a core 2 fail with +200 even when undervolt was 0 for that core (-5 0 -5 -10 -15 -10 curve) so I'm just going to ignore it since 12 rounds of regular p95 passed. CC is large fft, my cpu ran at 71 degrees during the test.

I played with it for a couple of hours, it was pretty depressing. For absolutely everything else it never once gave me a problem with my settings. But.. I am enjoying my 5600X for now. My system is so quiet. Even when it gets a bit busy they barely lift from idle. Unlike 5900X which breathes fire lol.

 

[oobymach]

I played with it for a couple of hours, it was pretty depressing. For absolutely everything else it never once gave me a problem with my settings. But.. I am enjoying my 5600X for now. My system is so quiet. Even when it gets a bit busy they barely lift from idle. Unlike 5900X which breathes fire lol.

I'm testing an extra -2 per core (-12 -7 -7 -12 -17 -12 curve) @ +150 and core 2 crashes again in CC. Dif in single core is 8-10 points in cpuz so not a huge drop vs +200 but CC still doesn't like it, I'm just going to go back to +200 and pretend CC isn't a thing.

 

[Taraquin]

I played with it for a couple of hours, it was pretty depressing. For absolutely everything else it never once gave me a problem with my settings. But.. I am enjoying my 5600X for now. My system is so quiet. Even when it gets a bit busy they barely lift from idle. Unlike 5900X which breathes fire lol.

Seems only 5600X is the quiet one of the 5XX0X-brotherhood for now. I have a bit of hope for 5700X and 5600, they should be good.

Or 5950X is possibly the best of cooler is good, lower allcore clocks might be the reason:

 

[mattferris]

I haven't gotten above 82C on prime95 with just Small FFT with AVX on, cinebenchr23 is 10C lower it's also like 300mhz higher lol
My current issues seems to be my VRM loop 1 is hotter then VRM loop 2 whey they should mirror each other. I wonder if it just has bad contact.
Just now change my EDC from 125 to 85 = better Cinebenchr23 score both single and Multi. guess I'll try to lower the wattage and see what it gets.

Prime95 is a weird load compared to Cinebench or other more traditional workloads. The heavy AVX Prime95 computations will slam you into your TDC limit before PPT or EDC (unless you have PPT and EDC lowered sufficiently). This can lead to weird outcomes like you saw above.

 

[DemonicRyzen666]

Prime95 is a weird load compared to Cinebench or other more traditional workloads. The heavy AVX Prime95 computations will slam you into your TDC limit before PPT or EDC (unless you have PPT and EDC lowered sufficiently). This can lead to weird outcomes like you saw above.

I've been doing a little tinkering I let the motherboard set all the setting by just hitting the + button in the bios from auto

this is what it set

PPT 500W
TDC 143a
EDC193a

Then I ran Prime95/cinbenchr23 it only used this % of those settings on HWinfo.
500w 25%
143a 56.5%
193a 59.9%

which then looks like this
ppt 125w
TDC 78a
EDC 115a

I did notice that even if I switch the TDC with EDC
PPT 125w
TDC115
EDC 78

PPT won't make it up past 87.5 watts then. Think I'm going to test some static overclocks now

 

[freeagent]

Using 200/140/180 with my 5600X capped at +150 can put out over 140w PPT using Linpack Xtreme. The most I saw was just over 150w PPT on 5600X. Static OC should be easy on that CPU.