5950x and PBO overheated something or busted custom loop?

[RJARRRPCGP]

I mean, start by checking the pads (and the required thickness)
With pads and a block on em, even cheap VRMs should be fine... it's not a 5950x with PBO can blast to 250W



Oof you were not joking, it really does have bad VRMs.
This video does have some really wobbly camera work later on, if you get motion sickness

(The intro has the advertised specs, around 4 minutes he finds out what it really has)

Outright false marketing and claiming different VRM vendors
Claims 'Digi Power 12 phase VRM, 45A choke'
(They now claim intersil 12 phase on the website, but teardowns prove its 4)

Turns out it's a 4 phase from what he describes as a fireworks factory :/

On another video, I don't remember hearing about the B550 PG Velocita having VRM problems, OTOH. So, it better not die on me with a Ryzen 7 5800X!

 

[Mussels]

On another video, I don't remember hearing about the B550 PG Velocita having VRM problems, OTOH. So, it better not die on me with a Ryzen 7 5800X!

It's hard to find information on some of these boards as many motherboard vendors released outright lies about VRM performance, sometimes as bad as what brand they were, others saying it's 12 phases when its's really 3 or 4 due to doubles and some not even being connected (buildzoids videos are amazing) and then sometimes they sneakily released newer revisions, and then many reviewers tested lower wattage CPUs with PBO off or tested all core OC's with locked voltages on ryzen 3000 where this tended to use *LESS* wattage than stock or PBO - few tested Zen 3 CPU's, even less with PBO enabled

'B550 PG Velocita' seems to be on the decent side, almost no reviews or comments on the VRM's other than saying it's the same as the B550 steel legend

12 phase sounds good, 50A is bad, double explains a lot - comes out equal to 6 phase 90A you see on newer mid-range boards by my estimates.
These guys noted its a 6+2 controller so it cant possibly be 12 actual phases, but companies loved the misleading advertising back then.

This should be enough even for CPUs like the 16-core Ryzen 9 3950X.

This is a whole system wattage not CPU alone, but you can see the zen3 CPU's use *less* than Zen 2 counterparts, as long as PBO is disabled - the moment you enable it to use one feature like curve undervolt, make sure to manually enter PBO limits so it doesnt blast off to 250W+

This is from a YT video, where a user enabled PBO and fired up cinebench to compare results - but it shows the max wattage leaping from ~140W to 240W and that's how all but the very best boards get overwhelmed

 

[freeagent]

Yeah with PBO these things have no problem sitting at 200w.

 

[RJARRRPCGP]

Whew! I guess I probably won't suddenly find it off unexpectedly and super dead!

 

[Mussels]

Yeah with PBO these things have no problem sitting at 200w.

View attachment 294396

200W seems high to me, but you've got twice the surface contact for cooling i do with single CCX chips - so if i can handle 120W, you can handle 240W i guess.

Users just gotta make sure they know what wattages their board can handle, and ensure they dont blast past it. It won't usually result in dead hardware, but instead stuttering and performance issues.

 

[Dr. Dro]

Well my experience with the 5950X I had until recently is that with AGESA newer than 1.2.0.3 C (last pre-5800X3D one) you can forget about PBO working correctly, it's just not going to, the CPU will hard limit at 142 A EDC no matter what setting you use. It's awful and AMD has no intention on fixing the blasted EDC bug. The performance loss is up to 10% on Cinebench R23 vs. pre-5800X3D AGESA.

This will limit your CPU's performance to the 170 to 200 W realm at best depending on workload, you just don't have more room to work with even if you do have a capable motherboard. The bug is caused by the 5800X3D's hard EDC limiter kicking in on the other Ryzen models as well. 142 A isn't the actual limit, it's the CPU's stock fused EDC, on 5600X this is even more severe at 90 A only. Increasing EDC or attempting to push past this will force the reliability algorithm to reduce the clock speed, so since the 5800X3D released the other Ryzen models can no longer draw more current than they are configured to do so from the factory. If CPU fused EDC value is surpassed, it will apply a VID_LIMIT on SMU and reduce clocks and voltage until EDC limit is met.

Check thread on Reddit:

https://www.reddit.com/r/Amd/comments/vecyn6
No one cared and no one continues to care, so this circus greatly contributed to me selling it and purchasing my i9-13900KS.

 

[freeagent]

200W seems high to me, but you've got twice the surface contact for cooling i do with single CCX chips - so if i can handle 120W, you can handle 240W i guess.

Users just gotta make sure they know what wattages their board can handle, and ensure they dont blast past it. It won't usually result in dead hardware, but instead stuttering and performance issues.

Its a bit high, but not as high as 240

I have seen my X3D do its 143w PPT, and with PBO I think my 5600X is in that neighborhood too, but its gotta be running a high fclk/mclk to do it.

so since the 5800X3D released the other Ryzen models can no longer draw more current than they are configured to do so from the factory.

Old AGESA ftw

 

[Dr. Dro]

Its a bit high, but not as high as 240

I have seen my X3D do its 143w PPT, and with PBO I think my 5600X is in that neighborhood too, but its gotta be running a high fclk/mclk to do it.

Old AGESA ftw

If you didn't mind the fTPM stutter, constant USB dropouts, and routine memory training failures, all of which I was directly hit by.

Honestly, my experience with it only really improved with AGESA 1.2.0.7, but at that point I had to choose between the monumental amount of issues I had vs. only 90%ish of my CPUs former potential, I chose to sell it all and start anew. Wasn't an easy choice, I had to endure 4 months of using my laptop only to get it done... and spent more than I initially intended to, but it was worth it.

 

[A Computer Guy]

It's almost that time again. 1yr fluid change pending as well as some additional build modifications.

• Going to see if I can successfully cram xspc 360 thin rad in the side compartment.
  • Not sure how I want to reroute the tubes yet.
  • Also not sure if I want to flip it to exhaust or keep it intake.
• Possibly adding more quick disconnects.
• Boot drive SSD upgrade.
• D5 swap on the distro plate to see if the distroplate pump was just noisy.
• Perhaps some padding around the distroplate to dampen resonance from the pump.
  • Materials suggestions welcome. It's got to be thin <2mm to clear my top rad.
• Possible teardown and cleaning of distroplate. I can see a bit of residue on gaskets.
  • Not looking forward to the extra time this will take but it's the price you got to pay.
• Snagged a velocity 2 CPU block on sale! It's going in.
• Trying new Thermal Paste!
• CPU swap with my backup pc to see if the Taichi can drive my ram sticks with 128GB ram properly or not with newest UEFI/BIOS.
  • Questioning if my CPU was damaged from before limiting my ram capacity. Going to find out.
• Fix distroplate rgb wiring run with rgb extension cable.
• Hopefully I can get the disassembly/assembly done in two days.

other projects

• Some recent breakthrough with my ECC was able to do DDR4-3600 finally but I haven't gotten around to stability testing. If 128GB doesn't work out at DDR4-3200 then 64-GB at DDR4-3600 will be my goal and I'll stick the remaining 64GB in my backup pc.

Like if you want to see new teardown and rebuild photos!

 

[tabascosauz]

It's hard to find information on some of these boards as many motherboard vendors released outright lies about VRM performance, sometimes as bad as what brand they were, others saying it's 12 phases when its's really 3 or 4 due to doubles and some not even being connected (buildzoids videos are amazing) and then sometimes they sneakily released newer revisions, and then many reviewers tested lower wattage CPUs with PBO off or tested all core OC's with locked voltages on ryzen 3000 where this tended to use *LESS* wattage than stock or PBO - few tested Zen 3 CPU's, even less with PBO enabled

'B550 PG Velocita' seems to be on the decent side, almost no reviews or comments on the VRM's other than saying it's the same as the B550 steel legend

View attachment 294392

12 phase sounds good, 50A is bad, double explains a lot - comes out equal to 6 phase 90A you see on newer mid-range boards by my estimates.
These guys noted its a 6+2 controller so it cant possibly be 12 actual phases, but companies loved the misleading advertising back then.

This is a whole system wattage not CPU alone, but you can see the zen3 CPU's use *less* than Zen 2 counterparts, as long as PBO is disabled - the moment you enable it to use one feature like curve undervolt, make sure to manually enter PBO limits so it doesnt blast off to 250W+
View attachment 294393

This is from a YT video, where a user enabled PBO and fired up cinebench to compare results - but it shows the max wattage leaping from ~140W to 240W and that's how all but the very best boards get overwhelmed

It is technically not 12 phases but in the realm of thermal performance there is no difference. 12 Vishay 50A is still 12 power stages, it can be 12 phases out of an XDPE, 6 doubled from an IR35201 or RAA229004, or 4 tripled from a Richtek and the thermal difference will be negligible. Performance-wise is where the controller differences could potentially appear, but Zen CPUs are so dynamic these days that it's hard to single out any variable including PWM performance.

Granted, board implementation differs but 12 phase DrMOS is not an issue. 8 x 50A is still overwhelmingly fine at 142W, but bad heatsinks can make/break an 8-phase at PBO ~180W, though probably still not reaching 110C.

PG Velocita has a leg up on the Steel Legend as Steel Legend is a 4-layer board which hampers PCB heat dissipation a bit.

It's almost that time again. 1yr fluid change pending as well as some additional build modifications.

• Going to see if I can successfully cram xspc 360 thin rad in the side compartment.
  • Not sure how I want to reroute the tubes yet.
  • Also not sure if I want to flip it to exhaust or keep it intake.
• Possibly adding more quick disconnects.
• Boot drive SSD upgrade.
• D5 swap on the distro plate to see if the distroplate pump was just noisy.
• Perhaps some padding around the distroplate to dampen resonance from the pump.
  • Materials suggestions welcome. It's got to be thin <2mm to clear my top rad.
• Possible teardown and cleaning of distroplate. I can see a bit of residue on gaskets.
  • Not looking forward to the extra time this will take but it's the price you got to pay.
• Snagged a velocity 2 CPU block on sale! It's going in.
• Trying new Thermal Paste!
• CPU swap with my backup pc to see if the Taichi can drive my ram sticks with 128GB ram properly or not with newest UEFI/BIOS.
  • Questioning if my CPU was damaged from before limiting my ram capacity. Going to find out.
• Fix distroplate rgb wiring run with rgb extension cable.
• Hopefully I can get the disassembly/assembly done in two days.

other projects

• Some recent breakthrough with my ECC was able to do DDR4-3600 finally but I haven't gotten around to stability testing. If 128GB doesn't work out at DDR4-3200 then 64-GB at DDR4-3600 will be my goal and I'll stick the remaining 64GB in my backup pc.

Like if you want to see new teardown and rebuild photos!

I do wanna see this, because I've been entertaining putting my loop back together. Wanna find out how you clean your parts, because all of mine need a good wash after sitting there dry for a year (Foundation AM4, heatkiller res and DDC, XT45 rad).

 

[Dr. Dro]

Like if you want to see new teardown and rebuild photos!

Good stuff, let's see you give your machine some love

 

[Mussels]

Well my experience with the 5950X I had until recently is that with AGESA newer than 1.2.0.3 C (last pre-5800X3D one) you can forget about PBO working correctly, it's just not going to, the CPU will hard limit at 142 A EDC no matter what setting you use. It's awful and AMD has no intention on fixing the blasted EDC bug. The performance loss is up to 10% on Cinebench R23 vs. pre-5800X3D AGESA.

This will limit your CPU's performance to the 170 to 200 W realm at best depending on workload, you just don't have more room to work with even if you do have a capable motherboard. The bug is caused by the 5800X3D's hard EDC limiter kicking in on the other Ryzen models as well. 142 A isn't the actual limit, it's the CPU's stock fused EDC, on 5600X this is even more severe at 90 A only. Increasing EDC or attempting to push past this will force the reliability algorithm to reduce the clock speed, so since the 5800X3D released the other Ryzen models can no longer draw more current than they are configured to do so from the factory. If CPU fused EDC value is surpassed, it will apply a VID_LIMIT on SMU and reduce clocks and voltage until EDC limit is met.

Check thread on Reddit:

https://www.reddit.com/r/Amd/comments/vecyn6
No one cared and no one continues to care, so this circus greatly contributed to me selling it and purchasing my i9-13900KS.

That reddit thread mentions using PBOtuner software bypasses the issue, if that helps you


This may factor into the issues people are having with stuttering setups on AM4 in a few other threads, as many of them enable PBO at maximum values and never look back at them

 

[A Computer Guy]

This may factor into the issues people are having with stuttering setups on AM4 in a few other threads, as many of them enable PBO at maximum values and never look back at them

I have an x470 ITX recently setting up as a test bench for some reason had PBO enabled by default with my 3800x with some wonky random values. I was having terrible USB and mouse stuttering input, although the PBO values weren't sky high, after I reset to defaults and turned PBO off things were normal. In ECO Mode (65w) I also discovered with some quick benching CPUz and CB15 had very little performance loss on my 3800x but I have a wraith stealth cooler strapped to it at the moment which is probably a somewhat limiting factor.

 

[Dr. Dro]

That reddit thread mentions using PBOtuner software bypasses the issue, if that helps you


This may factor into the issues people are having with stuttering setups on AM4 in a few other threads, as many of them enable PBO at maximum values and never look back at them

I no longer have the CPU, but that would mostly account for a CPU that could go far in curve optimizer, no? My 5950X was either a stinker or clocked high out of the box, since all I could do was -2 all core, and that was it.

 

[Mussels]

It is technically not 12 phases but in the realm of thermal performance there is no difference. 12 Vishay 50A is still 12 power stages, it can be 12 phases out of an XDPE, 6 doubled from an IR35201 or RAA229004, or 4 tripled from a Richtek and the thermal difference will be negligible. Performance-wise is where the controller differences could potentially appear, but Zen CPUs are so dynamic these days that it's hard to single out any variable including PWM performance.

Granted, board implementation differs but 12 phase DrMOS is not an issue. 8 x 50A is still overwhelmingly fine at 142W, but bad heatsinks can make/break an 8-phase at PBO ~180W, though probably still not reaching 110C.

PG Velocita has a leg up on the Steel Legend as Steel Legend is a 4-layer board which hampers PCB heat dissipation a bit.



I do wanna see this, because I've been entertaining putting my loop back together. Wanna find out how you clean your parts, because all of mine need a good wash after sitting there dry for a year (Foundation AM4, heatkiller res and DDC, XT45 rad).

It's not my area of expertise, but what tended to come up in buildzoids videos was that these phases weren't implemented correctly - so while they may have had 12 of them, the way they were wired up physically prevented them working at their best.

While the cheaping out and crap designs started pre-covid, they got worse with slap-in replacements for out of stock parts due to covid delays a lot of the time, and they didn't re-design boards for the cheaper components and often they simply didnt work as well as they could have, he summarised it as basically being less efficient due to the secondary parts they combined them with (the controller chips, usually)

You mentioned 110C but a lot of these boards VRM's arent rated for that high either, some as low as 80C or 90C were on boards with no heatsinks

 

[tabascosauz]

It's not my area of expertise, but what tended to come up in buildzoids videos was that these phases weren't implemented correctly - so while they may have had 12 of them, the way they were wired up physically prevented them working at their best.

While the cheaping out and crap designs started pre-covid, they got worse with slap-in replacements for out of stock parts due to covid delays a lot of the time, and they didn't re-design boards for the cheaper components and often they simply didnt work as well as they could have, he summarised it as basically being less efficient due to the secondary parts they combined them with (the controller chips, usually)

You mentioned 110C but a lot of these boards VRM's arent rated for that high either, some as low as 80C or 90C were on boards with no heatsinks

ASRock has never used anything on its DrMOS AM4 boards other than variants of Vishay 50A parts, shortage or not. Except for the dual N-fets on their B450 ITX and possibly some PowIR in earlier gens. You might be thinking Gigabyte revising its X570/B550s to omit the ISL6617 doublers due to shortages. At the end of the day the differences were negligible and the pure doubled boards weren't really any different.

I don't think I've ever heard bz say on any board breakdown that certain phases aren't wired up correctly. Only one that stood out as being horrible was the Biostar board and that was due to lack of layers and/or extraordinarily bad power plane design that ruined the performance of the six ISL99390s.

There are obviously minor benefits to having 12 true phases, and some minor performance and efficiency differences between doubling with doublers (ASRock) and without (Asus), but in terms of thermal and current handling 12 Vishays is 12 Vishays. ASRock at the low end prefers to use cheap and dumb analog controllers from UPI, but again irrelevant to current handling.

All Vishay, Infineon and IR datasheets I've seen are at least 110C junction rated, it's pretty standard for DrMOS and SPS. Vishay is 150C. Whether there is a heatsink or not (very rarely a choice if it's DrMOS or SPS except on super low end AM5/LGA1700) doesn't change what the part is specced for. Board vendor may choose to throttle the entire VRM aggressively based on temp (ASRock), but that's a different issue. All the common DrMOS and SPS can take a lot more punishment than most people think.

 

[Mussels]

I was definitely being generic there, and not specifically about asrock
You specifically also said "they never used anything but..." and then gave the exceptions for when they did use different parts

You'd have to watch buildzoids VRM breakdowns, he goes into the math of it all and there's usually no simple summaries without watching the entire videos

 

[tabascosauz]

You'd have to watch buildzoids VRM breakdowns, he goes into the math of it all and there's usually no simple summaries without watching the entire videos

What makes you think I don't? the math in the PCB breakdowns is a good general reference but not taken literally.

And the AM4 reality has always been that once you get to 8 x 50A Vcore with a decent heatsink (barring astounding unforeseeable fuckups like B550GTQ), everything else is nitpicking because it can easily handle what AM4 can throw at it. And knowing that, hell yeah I agree with his rambling - once you cross that threshold of 8x50A, there's nothing really left to talk about.

As to the phases "not connected" to the VRM, I don't recall hearing about that for any X570/B550 board, but do link a video if you come across it, to jog memory. Closest I can think of to that idea is that a lot of ISL6617-equipped boards aren't necessarily programmed to take advantage of its full doubling/current balancing(?) features, but again that does not affect a VRM's current handling capability, just kinda a waste of money for doublers.

 

[A Computer Guy]

So next weekend is looking good for starting the 1yr maintenance. I took some time to outline my process hopefully I'm not forgetting something but this should also help in planning when and where to take photos. In the process I managed to document some outstanding questions some of which I will just have to figure out during the rebuild.

It would seem I can't paste my bulleted list properly...snag-it to the rescue with scrolling window capture...

Current configuration. I turned on the rainbow puke so the photo would take better in the dark.

Questions

• Suggestions for sound dampening materials (<2mm)
• QDC hookup orientation for least flow restriction? (I came across a suggestion somewhere on TPU but can't find it anymore.)

 

[Count von Schwalbe]

Suggestions for sound dampening materials (<2mm)

For vibration isolation or sound absorption? If the former, go for neoprene sheet/washers. Remember that vibrations can be transmitted through screws if they are not fully isolated from the vibrating part as well.

 

[A Computer Guy]

For vibration isolation or sound absorption? If the former, go for neoprene sheet/washers. Remember that vibrations can be transmitted through screws if they are not fully isolated from the vibrating part as well.

vibration isolation. if I recall correctly the other side of the distroplate mounting was countersunk machine thread screws to the front of the case. (edit) yea I just confirmed it. There is no way I'd be able to put a washer on that side or the glass likely won't fit back into place. Maybe some kind of goo might work for the screws? Maybe if I use some plastidip spray on the screw that might be ok or LocTite https://www.amazon.com/Loctite-Heavy-Duty-Threadlocker-Single/dp/B000I1RSNS/ref=asc_df_B000I1RSNS/ might be a better option for the screws.

 

[Mussels]

So next weekend is looking good for starting the 1yr maintenance. I took some time to outline my process hopefully I'm not forgetting something but this should also help in planning when and where to take photos. In the process I managed to document some outstanding questions some of which I will just have to figure out during the rebuild.

It would seem I can't paste my bulleted list properly...snag-it to the rescue with scrolling window capture...

View attachment 295654


Current configuration. I turned on the rainbow puke so the photo would take better in the dark.

View attachment 295650

Questions

• Suggestions for sound dampening materials (<2mm)
• QDC hookup orientation for least flow restriction? (I came across a suggestion somewhere on TPU but can't find it anymore.)

I like my QDC's in my build from Alphacool, fairly compact and easy to add/remove.

AFAIK vinyl is one of the best dampeners due to it adding weight and density while absorbing a lot of movement, while foam is used to absorb higher frequency noise - but that noise escapes any gaps, making it better to focus on removing the noise in the first place.
Sound absorbing materials turn the sound into physical vibration then spread it out, while sound blocking materials turn it into an echo that can escape elsewhere - which high frequency stuff loves to do. Often something to absorb that is better OUT of the PC, like something hanging from the wall behind it to stop it echoing to you.

What are you dealing with, fan noise, vibration, coil whine?

 

[Ferrum Master]

It looks good, but I am not fond of distro plates. It kinda is meant to reduce clutter, but it is full of spaghetti tubes. I have used hard extenders on tank and pump sections to get rid of needing to mount it anywhere. And the manometer is actually functional, I have two pumps, other is behind the board, and the main pump is around 20% PWM under no load, totally silent. I need a new main pump either way, it old and and rattling a bit.

 

[Count von Schwalbe]

vibration isolation. if I recall correctly the other side of the distroplate mounting was countersunk machine thread screws to the front of the case. (edit) yea I just confirmed it. There is no way I'd be able to put a washer on that side or the glass likely won't fit back into place. Maybe some kind of goo might work for the screws? Maybe if I use some plastidip spray on the screw that might be ok or LocTite https://www.amazon.com/Loctite-Heavy-Duty-Threadlocker-Single/dp/B000I1RSNS/ref=asc_df_B000I1RSNS/ might be a better option for the screws.

The usual method is to isolate the head of the screw from the vibrating part, then isolate the part from the mounting surface (in this instance it would be the case).

 

[A Computer Guy]

It looks good, but I am not fond of distro plates. It kinda is meant to reduce clutter, but it is full of spaghetti tubes. I have used hard extenders on tank and pump sections to get rid of needing to mount it anywhere. And the manometer is actually functional, I have two pumps, other is behind the board, and the main pump is around 20% PWM under no load, totally silent. I need a new main pump either way, it old and and rattling a bit.

Yea I didn't much care for the increase in tubes either however my original tube runs really didn't jive with how the distroplate was meant to be used. Going from 4 tubes to 6+1 was a bit of a shock. Now with the new rad +1 for a total of 8 although I am much happier with the primochill tubes that replaced the EK duraclear last time. I would have liked those ek matte black tubes but my fittings were the wrong size for that however the promochill black tubes have a nice shine that isn't obnoxious and looks really nice at night when my rig is completely red. I kinda hate to get rid of the distroplate since it was fairly expensive. If the pump noise / resonance issue persists I might go back to using my tube pump/res combo and simple tube runs. I've thought about getting the 011D EVO but it will only be bigger and heavier or perhaps 011D Air would be the smaller and better choice but then I have to get new rads.