help with undervolting 5800x3d

[Jacks.]

If you heard cracking then something isn’t right, take it apart and start over.

i got my hands on a d15 chromax today rented it from a shop to test with it and see in a few hours after work i will try again

ok got the cooler and here is the test result please tell me it's good

 

[Colddecked]

You can tell its working alot better. Score went up, max boost went up, temps went down.

 

[Jacks.]

this is with -30 pbo tuner do i need to do anything else ? stress test it with other programs ? or this is just enough

 

[freeagent]

Much better.

 

[Colddecked]

this is with -30 pbo tuner do i need to do anything else ? stress test it with other programs ? or this is just enough

Core cycler. Leave it on over night for a few nights. Mine has gone over 24hrs total on it so far.

 

[Jacks.]

Core cycler. Leave it on over night for a few nights. Mine has gone over 24hrs total on it so far.

you mean leaving cb r23 on for few nights ? or just the pc on ?
anyway here is how it ended -for now- i did get to the 15000 score in a test and then stressed it for 30 minutes with the room shut and window closed no ac working or ceiling fan the two test result are here

and this was my setting in bios 4006 then updated to 4201 to see if there is any difference
AMD CBS > NBIO > SMU > CPPC Enabled
AMD CBS > NBIO > SMU > CPPC Preferred Cores Disabled
AMD CBS > CPU > Global C-State Control Enabled
and i have a ram voltage down to 1.2v
now hardware wise i was right about lian li cooler screws there is something wrong with that after installing d15 and tightened the screws down it did stop and didn't want to move again as for the galahad that didn't happen at all it just kept going which is very odd as i said maybe i don't have an experience in overclocking or undervolting at all but i have built computers more that i can remember and don't think i saw that at all i literally heard the plastic bracket cracking and when i took it out i saw a little crack there maybe there is something faulty with the springs i don't know ,right now i stopped testing cause it's 4 in the morning and cause d15 isn't mine i have to return it tomorrow but now i know that the cpu is fine and the problem is the cooler i will be getting d15 in the next few weeks so if you see this thank you very much for helping me though this great community really glad to be a part of it and if you have more advises please tell me

 

[Zach_01]

I see PPT only goes as far as ~95W. Did you change PPT limit from stock? Isn’t it 142W stock like the other 105W TDP CPUs?

I should better ask what is the Power Reporting Daviation and PPT during the CB R23 multi run, because PPT alone doesn’t mean much.

 

[Jacks.]

I see PPT only goes as far as ~95W. Did you change PPT limit from stock? Isn’t it 142W stock like the other 105W TDP CPUs?

I should better ask what is the Power Reporting Daviation and PPT during the CB R23 multi run, because PPT alone doesn’t mean much.

no i didn't change ppt limits at all and what does this mean ? i know what ppt is but if it didn't change is this good or bad

 

[Zach_01]

I just wanted to know what is the total power consumption of the CPU.

If you’re having a Power Reporting Deviation value below or above 100% then what you see at PPT sensor is not true.

It’s an easy calculation to find the true value for total power.

For example if PRD is 80% and PPT is 90W during CB R23 MT then the power of the CPU package is:

90 / 0.8 = 112.5W

Note that PRD value has meaning only under full CPU load. In any other situation you just ignore it. Doesn’t mean anything and the value can be anything from 0% to several hundreds.

 

[tabascosauz]

i got my hands on a d15 chromax today rented it from a shop to test with it and see in a few hours after work i will try again

ok got the cooler and here is the test result please tell me it's good

That's pretty damn good and spot on. I guess it was the AIO, whatever the reason. Maybe the mounting just wasn't that great on your kit. When mounting waterblocks it can take a lot of work to tighten it down enough, unlike air coolers, and hard to tell if it's "good enough".

For comparison I'm also basically at the 15000 mark, about 76C at -30, about 100W for that score. Considering your ambient is so much higher (I think you said 29? Mine is usually 21-22 in the mornings), your D15 is really doing good work.

Case closed, I guess? Enjoy your setup, you've got a good thing going. If you really really want to take a screenshot for posterity, you can close some background stuff, disable Search service and internet, or run in Safe Mode - should easily get you past 15000, you're already maxed out at 4450.

With Noctua and most good air coolers, if the screws stop, they stop. No need to try and give it more force to see if it goes further, it gives you very clear feedback when it reaches the end.

I applaud your perseverance. It's very frustrating when trying to hunt down a cooling problem and mounting/remounting all day.

I see PPT only goes as far as ~95W. Did you change PPT limit from stock? Isn’t it 142W stock like the other 105W TDP CPUs?

I should better ask what is the Power Reporting Daviation and PPT during the CB R23 multi run, because PPT alone doesn’t mean much.

The X3D is a bit strange, it's rare (if even possible at stock) to see it max out 142W even if that's what you allow it to draw. Most people fall around 100-120W area stock in Cinebench, CO may or may not affect that. Personally I haven't found that much difference between about 100W and 120W, what matters is getting to 4450MHz all core effective (since 4450 is Fmax for all-core load). There are a lot of things about this CPU that go against conventional Zen 3 wisdom.

Beyond 120W temps really start to climb, and due to Fmax and basically score cap at 15k, no point going further. AMD set way too ambitious a PPT out of the box for these. For scores I'd let it go up to 100/65/90, but honestly the usual Eco limits work just as well if not chasing that last 1% of benchmark scores (88/60/90).

With a good cooler, if you can hit that clock target, 100W should be all you really need for 15k (after 15k the only real improvements in score will come from BCLK).

 

[Jacks.]

90 / 0.8 = 112.5W

Note that PRD value has meaning only under full CPU load. In any other situation you just ignore it. Doesn’t mean anything and the value can be anything from 0% to several hundreds

ok here you go first pic is with co applied

and this without co

these were during the test not after

That's pretty damn good and spot on. I guess it was the AIO, whatever the reason. Maybe the mounting just wasn't that great on your kit. When mounting waterblocks it can take a lot of work to tighten it down enough, unlike air coolers, and hard to tell if it's "good enough".

For comparison I'm also basically at the 15000 mark, about 76C at -30, about 100W for that score. Considering your ambient is so much higher (I think you said 29? Mine is usually 21-22 in the mornings), your D15 is really doing good work.

Case closed, I guess? Enjoy your setup, you've got a good thing going. If you really really want to take a screenshot for posterity, you can close some background stuff, disable Search service and internet, or run in Safe Mode - should easily get you past 15000, you're already maxed out at 4450.

With Noctua and most good air coolers, if the screws stop, they stop. No need to try and give it more force to see if it goes further, it gives you very clear feedback when it reaches the end.

I applaud your perseverance. It's very frustrating when trying to hunt down a cooling problem and mounting/remounting all day.

most of the thanks goes to you my friend you have been very very helpful and yeah the d15 is a very good cooler the only down side is it's size it will block the gpu latch and the first m.2 slot but if you want cooling then noctua got my back i will do further testing with the galahad on another system to make sure it's working or it's just one of those which got there fins clogged

 

[tabascosauz]

ok here you go first pic is with co applied

and this without co

these were during the test not after

most of the thanks goes to you my friend you have been very very helpful and yeah the d15 is a very good cooler the only down side is it's size it will block the gpu latch and the first m.2 slot but if you want cooling then noctua got my back i will do further testing with the galahad on another system to make sure it's working or it's just one of those which got there fins clogged

Glad you were able to sort it all out

Very interesting how the CO is working on your board. On both my boards it reduces temps and increases clocks, but has no effect on wattage by itself. Didn't reduce wattage for my 5900X, 5600G or 5700G either. But that's just the thing with Ryzen I guess, everyone's setup can work slightly differently out of the box.

I know the struggle very well over cooler size......the Impact's SO-DIMM.2 also makes it hard to reach GPU release latch, and no way to get SO-DIMM.2 out without removing either the C14S or GPU. But Noctua has good quality especially on the mounting hardware - the C14S alone I've mounted/remounted maybe 100 times, if not more. It feels every so slightly looser now in some aspects, but it still gets 100% of the performance.

I would be interested to see how the AIO turns out, now that there's no urgency to testing it.

 

[Mussels]

i got my hands on a d15 chromax today rented it from a shop to test with it and see in a few hours after work i will try again

ok got the cooler and here is the test result please tell me it's good View attachment 266228

We need to see that while the test is running, so we can see the clock speeds

Averages dont tell us the full story if you've had it idling - With it being open for an entire 1 minute and 41 seconds we cant really get much useful info from that


set it to a 10 minute + run and take the screenshot while it's still rendering, near the end

 

[Jacks.]

set it to a 10 minute + run and take the screenshot while it's still rendering, near the end

there is a pic while it's running just up there

 

[Mussels]

those results are running for 1 minute and 2 minutes

They're useless for anything other than 'current' values

 

[Zach_01]

The X3D is a bit strange, it's rare (if even possible at stock) to see it max out 142W even if that's what you allow it to draw. Most people fall around 100-120W area stock in Cinebench, CO may or may not affect that. Personally I haven't found that much difference between about 100W and 120W, what matters is getting to 4450MHz all core effective (since 4450 is Fmax for all-core load). There are a lot of things about this CPU that go against conventional Zen 3 wisdom.

Beyond 120W temps really start to climb, and due to Fmax and basically score cap at 15k, no point going further. AMD set way too ambitious a PPT out of the box for these. For scores I'd let it go up to 100/65/90, but honestly the usual Eco limits work just as well if not chasing that last 1% of benchmark scores (88/60/90).

With a good cooler, if you can hit that clock target, 100W should be all you really need for 15k (after 15k the only real improvements in score will come from BCLK).

Glad you were able to sort it all out

Very interesting how the CO is working on your board. On both my boards it reduces temps and increases clocks, but has no effect on wattage by itself. Didn't reduce wattage for my 5900X, 5600G or 5700G either. But that's just the thing with Ryzen I guess, everyone's setup can work slightly differently out of the box.

I know the struggle very well over cooler size......the Impact's SO-DIMM.2 also makes it hard to reach GPU release latch, and no way to get SO-DIMM.2 out without removing either the C14S or GPU. But Noctua has good quality especially on the mounting hardware - the C14S alone I've mounted/remounted maybe 100 times, if not more. It feels every so slightly looser now in some aspects, but it still gets 100% of the performance.

I would be interested to see how the AIO turns out, now that there's no urgency to testing it.

I guess it is unlike other 5000series and these limits are only "just names" apparently, and most likely it behaves like that because of the hard cap at clocks. So -CO does real undervolt since/if the CPU already hits the cap with out -CO.
Hence the lower wattage between the 2 runs w/ and w/o -CO.

You know the following but I'll just say it for the record
Every other 5000 will just keep boosting at +150~350MHz beyond its advertised speeds when/if possible so saving watts (when -CO is applied) is going down the drain, unless there is a hard-cap setting on power/current limit by the user.

ok here you go first pic is with co applied
View attachment 266243
and this without co
View attachment 266244
these were during the test not after

most of the thanks goes to you my friend you have been very very helpful and yeah the d15 is a very good cooler the only down side is it's size it will block the gpu latch and the first m.2 slot but if you want cooling then noctua got my back i will do further testing with the galahad on another system to make sure it's working or it's just one of those which got there fins clogged

Probably this is as close as you can get to best thermals without braking the bank while having the most perf out of this CPU.

If I didn't know about the special core die configuration of the X3D I would say that 75~80C at 95~100W PPT is not all that good but this is not the case here.
Given the fact that there is an extra layer of cache on top of the "typical" Zen3 core die that thickens things under the hood I think we can say "the job is done".

 

[tabascosauz]

I guess it is unlike other 5000series and these limits are only "just names" apparently, and most likely it behaves like that because of the hard cap at clocks. So -CO does real undervolt since/if the CPU already hits the cap with out -CO.
Hence the lower wattage between the 2 runs w/ and w/o -CO.

You know the following but I'll just say it for the record
Every other 5000 will just keep boosting at +150~350MHz beyond its advertised speeds when/if possible so saving watts (when -CO is applied) is going down the drain, unless there is a hard-cap setting on power/current limit by the user.

If I didn't know about the special core die configuration of the X3D I would say that 75~80C at 95~100W PPT is not all that good but this is not the case here.
Given the fact that there is an extra layer of cache on top of the "typical" Zen3 core die that thickens things under the hood I think we can say "the job is done".

5800X3D generally can run even hotter than 5800X at same Vcore and power. It's pretty good for OP since their ambient is like 30C. iirc the actual cache sits over cache (middle), the core part is covered by a silicon spacer that apparently doesn't work great for thermals.

I wouldn't go into it with the expectation that CO will reduce power consumption on X3D, it is just very different depending on setup. I can adjust on the fly with PBO2 Tuner and both -0 and -30 will both land me around 120W package stock (obviously, one is at 76C and one is at 90C).

AGESA is very messy for this CPU. We can only use 1206 or 1207 BIOSes, but so far we have vendors forgetting to hide PBO menus, vendors killing performance on 1207 (mine drops from 4380 to 4250 in R23 on 1207, thankfully performance otherwise is great), and apparently 1206 can't even be modded because entire menu like CBS can't even be found

those results are running for 1 minute and 2 minutes

They're useless for anything other than 'current' values

How's it useless? R23 clocks shouldn't be all that variable over the course of a single run on any Zen 3, drop maybe 50-100Mhz at most. Unless the BIOS team at your board manufacturer was smoking a blunt at work and screwed the pooch royally

What would a 10 minute run solve, Cinebench clocks has a big dip in the few second gap between runs, I have a hard time believing Average column is somehow better than Mark 1 Eyeball watching HWInfo during the test

As long as scores are about right and snapshot core clock is close to effective, what's the issue

 

[Jacks.]

those results are running for 1 minute and 2 minutes

They're useless for anything other than 'current' values

here you go 10 minutes run

 

[Zach_01]

here you go 10 minutes run
View attachment 266253

If you really want to do this right you have to do the following

1. Start the R23 MT
2. When squares start to fill wait 3~4sec and click reset (clock) button on HWiNFO sensors window
3. Take screenshot ~10sec before last run ends

This way you will capture the "Average" effective clocks of all cores and only during the run, and basically every other CPU sensor.

The 10min R23 run is a bit tricky for metrics as its not a continuous one but rather a cycling one with very small pauses in between. Who ever tries to interpret it will have to consider this.
I used it too though for CPU performance and boosting evaluation as I dont like burn/stress tests much.

FYI,
Core VID voltage values are not anything useful to see as these are just core requests and not what the core voltage really is.

 

[Jacks.]

If you really want to do this right you have to do the following

i got what you said and did it before i did many testing before posting about the d15 and got the same result didn't dip below 4.4 i just can't repeat now cause i have to return the cooler

 

[Mussels]

here you go 10 minutes run
View attachment 266253

That's at full speed.

No throttling, no clock stretching.
97-99W means the undervolt is definitely working, you're not going to get any better than this.

If your scores are lower than you expect, you've got software using those CPU cycles.

i got what you said and did it before i did many testing before posting about the d15 and got the same result didn't dip below 4.4 i just can't repeat now cause i have to return the cooler

They run 4.45 but seeing 4.41 like you did is normal as that's just the result of say 99.9MHz being measured and rounding down

 

[Taraquin]

That's at full speed.

No throttling, no clock stretching.
97-99W means the undervolt is definitely working, you're not going to get any better than this.

If your scores are lower than you expect, you've got software using those CPU cycles.


They run 4.45 but seeing 4.41 like you did is normal as that's just the result of say 99.9MHz being measured and rounding down

Turn off spread spectrum, that can improve MHz stability

 

[Jacks.]

ok so for everyone who still interested i finally got the cooler nh-d15 today and everything went as expected the scores is the same as the pic above that end that problem so thank you all for helping

 

[Mussels]

ok so for everyone who still interested i finally got the cooler nh-d15 today and everything went as expected the scores is the same as the pic above that end that problem so thank you all for helping

Whats the temps like now?

D15 should be about as good as you can get

 

[wolf]

Hijacking a little based on source content.

I have successfully done the in windows PBO2 undervolt with my X3D, -22 all core allows sustained clocks of 4450mhz all core in games, don't really feel the need to push further as it's rock stable, lower temps and not dropping clocks at all in game.

My question is, I feel like someone has previously mentioned it may be possible to find or request a custom version of my motherboard bios to allow PBO2 UV with the X3D, anyone here know who or where I would ask for that?