5950x and PBO overheated something or busted custom loop?

[A Computer Guy]

I was gonna say a D5 is rather quiet so that distro must create a bit of noise.

Yes that darn metal plate in front of the pump for instance reverbs like a loudspeaker. (I'm dramatizing of course but it conducts a fair bit of noise if it's pointed in your direction.)
I managed to pad the distroplate from the case which really helped but I think I might try some rubber spray on the frame in next time I do a complete teardown since the fix is already very tight.

I've got a 3900x in a x370 board... it's like hanging on for dear life that 3900x is bone stock. Anymore and that board will probably pop.

I have some suspect vrms in two other Asrock ITX (b450 and x470) boards so those are sticking to 8 cores max.

 

[Mussels]

to sum this thread up Way to much current on a so-so motherboard = melted stuffs

Doesn't help when the motherboard lies about it's components, 12 phase when it's 4 phase and '45A chokes' when the VRMs struggle to output 15A

was gonna say a D5 is rather quiet so that distro must create a bit of noise.

They can vibrate if they're mounted poorly, but otherwise all of mine have been dead silent at 100%
I have read of issues with some stock rubber gaskets being crappy and causing some vibration

Ahah - the 45A on the asrock board is the chokes and not the VRMs
They're putting that measurement where other brands put the MOSFETs ratings, so I made the mistake they wanted us to assume - that it's rated to 45A (when the chokes are, but the FETs are 12-15A)

 

[phill]

I've got a 3900x in a x370 board... it's like hanging on for dear life that 3900x is bone stock. Anymore and that board will probably pop.

I've two similar setups I used for WCG, I use the Asus Hero 6 boards. Ones on air and the other has an EVGA 280mm AIO on it, I have been impressed with the air cooler on the other board, it does very well I do run the CPUs undervolted though, seems to help no end

I've enjoyed catching up on the thread, so I hope everything is now up and running??

 

[A Computer Guy]

I've two similar setups I used for WCG, I use the Asus Hero 6 boards. Ones on air and the other has an EVGA 280mm AIO on it, I have been impressed with the air cooler on the other board, it does very well I do run the CPUs undervolted though, seems to help no end

I've enjoyed catching up on the thread, so I hope everything is now up and running??

Yes its working well although for some reason yesterday I lost connectivity to all my USB ports.

I'm just trying to wrap things up a bit to with tidbits that might be interesting to conclude the thread.

1) Does 128GB ram work ok on the new cpu? (was the old cpu potentially damaged after all?) (I'm waiting for the bios to come out of beta)
2) Revisiting PBO
3) At what point did I enter the danger zone with the old board? (a potentially new follow-up with the same model board I got at ebay)

 

[Jackanton]

Explaining AMD Ryzen Precision Boost Overdrive (PBO), AutoOC, & Benchmarks | GamersNexus

stub With the launch of the Ryzen 3000 series processors, we’ve noticed a distinct confusion among readers and viewers when it comes to the phrases “Precision Boost 2,” “XFR,” “Precision Boost Overdrive,” which is different from Precision Boost, and “AutoOC.” There is also a lot of confusion...

www.gamersnexus.net

Running PBO is probably what killed your system on top of trying to overclock your ram, maybe?

 

[A Computer Guy]

Corsairs commander pro works really well, you can set fan speeds static that it remembers and has 4x temp probes
View attachment 312908

I still have one that works but I don't want to tear down my system to borrow it.

The new core XT variant only has 2 temp probes, but its easier to get.


View attachment 312909


They work fine in HWinfo as long as you don't have iCue installed.

Yep. That was a pleasant surprise when I found that out too.

Since the commander pro is pretty hard to get these days. I started considering something like this for measuring temps although ideally having a second pc recording HWiNFO measurements would be nice.

Amazon.com

Also I need to find the right tool for measuring the power draw at the EPS connectors as you describe.


My first ever ebay package finally arrived today so I slapped it together with my leftover parts (Ryzen 2700, 32GB ECC, Quadro K600, Intel SATA SSD, and old iBuyPower case) to test that it actually works.
I used and old HP power/hdd-led front panel part since it's fast and easy to plug in.

I discovered some other interesting things about this old board when I tried to pop in my Asus M.2 card to see if x4x4x4x4 bifurcation would work.
The board won't post if the PCIe is not configured in x16 or x8x8 mode and the Quadro K600 will work in the open ended 1x slots at the top. x4 cards will likely fit in the 1x slots at the bottom but x16 like the quadro will bump into the button battery.

 

[Mussels]

Yes its working well although for some reason yesterday I lost connectivity to all my USB ports

I've seen this occur with low SoC voltage (x370) and from low VDDG (B450) - I think it was VDDG IO

I'm having a brain fart right now which one was related to the CPU ports dropping and which was related to chipset, in my case it seemed tied to the motherboard specifically - swapping CPUs and running it at higher ram speed still needed the same voltage(s) changed, just that the boards choice of when to raise them was weird (It may raise at 3400+, but not reach past 2667 with stock/auto voltages, since they don't go up in time)

 

[A Computer Guy]

So my UEFI/BIOS update finally came out with AMD AM4 AGESA Combo V2 PI 1.2.0.B and I had some time to retry 128GB ram which is one of the 3 last things on my list to complete this thread.
This time I opted to use genuine Micron memory and took out the Nemix ram. Still 32GB per stick DDR4-3200 EUDIMM 1.2v CL22 MTA18ASF4G72AZ-3G2R according to the label.

Comparing ZenTimings: Micron on the left , Nemix on the right. The timings are nearly identical however the Micron sticks have better tRFC.

Memtest86 already passed 1 round (which I expected) but the big test is if long idle times (typically overnight) cause spontaneous reboot.
To recap I'm using the swapped out undamaged CPU now with Micron branded memory. Previously was using the potentially damaged CPU and Nemix branded (micron) memory.

I'll have to review my 128GB thread for info that might be useful and do some testing with OCCT.

I have my event viewer already setup to filter log WHEA errors. It even shows the last error I had which I think was my last attempt with the 128GB Nemix ram.
Crossing my fingers. Hopefully things will go right this time.

 

[Mussels]

AMD AM4 AGESA Combo V2 PI 1.2.0.B

It's been so long since A came out, any idea what they fixed on this one?

Oh and if that generic memory turns out to be happy with the same TRFC, don't forget you can customise their SPD/XMP values. You can literally copy from the Micron to the Nemix, or edit at will.
Creating custom XMP profile for DDR4 | TechPowerUp Forums



Oh and dont forget SoC voltage - 128GB wont be stable without raising it, 8 ranks is only certified to 2667 on zen3.
Yours is at an odd value of 0.9875v - that's like a stock with LLC active, XMP should have raised that to 1.09v at least. 1.15v to 1.20v will likely be where the idle instability goes away.

 

[A Computer Guy]

It's been so long since A came out, any idea what they fixed on this one?

I think it was security fixes but if like anything else in software there maybe undocumented or non-public facing fixes.

Oh and if that generic memory turns out to be happy with the same TRFC,

I'm not sure what you mean here. I suspect the tRFC's must be different because the Micron and Nemix sticks use different Micron IC's. I know at 1.35v I can drive the tRFC down on the Nemix just fine but not super tight. Luckily I posted some of that Nemix history here so I can backtrack and do a little digging to find that info again.

don't forget you can customise their SPD/XMP values. You can literally copy from the Micron to the Nemix, or edit at will.
Creating custom XMP profile for DDR4 | TechPowerUp Forums

Yea that seems like an interesting prospect especially for the Nemix which I am planning to split between my ITX systems if everything works out well. A downside potentially is if I make the XMP profile too tight for my current CPU it might not work well for a different CPU later that might have a weaker IMC.

Oh and dont forget SoC voltage - 128GB wont be stable without raising it, 8 ranks is only certified to 2667 on zen3.

Thanks for the reminder. I havent lost sight I'm playing with the silicon lottery but some readers may not know that.

Yours is at an odd value of 0.9875v - that's like a stock with LLC active, XMP should have raised that to 1.09v at least. 1.15v to 1.20v will likely be where the idle instability goes away.

I forgot LLC was even a thing since I never configure it from the default. The sticks don't have XMP values of course so I'll have to adjust the voltages manually.

I found some older notes I made. With the Nemix I couldn't post those 4 sticks at all unless I setup SOC, CLDO VDDP, VDDG CCD, and VDDG IOD voltages manually first before installing. Again that was with the potentially damaged CPU. Now with the replacement CPU and Micron modules I reset UEFI/BIOS voltages to default, installed all four, then everything posted and booted no problem which is very hopeful to say the least. Like you said however I probably should increase the SOC voltage a bit. I'll test at the current voltages first and see what happens. These new Micron dimms also seem to run cooler.

When I had the Nemix in I did run the gambit of SOC voltage to 1.20v and DRAM to 1.45v, and frequencies from low to high but no joy on those Nemix at 128GB. The only problem finally being idle instability whenever I was away (and not looking at the system) with the potentially damaged CPU. That same CPU failed to drive my Corsair DDR4-3600 in my gaming ITX about a week or two ago (since I swapped the cpu's) so now I am thinking the IMC was/is degraded but then again they were Corsair sticks (shrug, the only non-ECC ram I'm running in my systems except the DeskMini) more testing needed I guess.

(edit...1 day later)

No spontaneous reboots overnight. I still have to do stress and heat testing tonight.

 

[Mussels]

I'm not sure what you mean here.

I meant if you want to keep using em together, mod the SPD/XMP values once you know what works

The sticks don't have XMP values of course

I missed that these were JEDEC 3200.

The big downside to that is that BIOSes tend to adjust voltages based on XMP being enabled, and without that they wont clock nearly as high

 

[A Computer Guy]

I meant if you want to keep using em together, mod the SPD/XMP values once you know what works
I missed that these were JEDEC 3200.

Yep just JEDEC ecc.

The big downside to that is that BIOSes tend to adjust voltages based on XMP being enabled, and without that they wont clock nearly as high

I think I understand what you mean but these are rated for DDR4-3200 at 1.2v.

So far they are clocking DDR4-3200 at the voltages previously indicated in the screenshot.
Survived 2 nights without spontaneous reboot and no WHEA errors, so I'm getting high confidence this is working and going to work going forward.
Got through some rounds of OCCT 30 minutes memory stress testing today and 2 days of work no issues.
I have to try power saving power plan and see if any stability issues occur including resume from sleep. (normally this PC runs 24/7 and never sleeps)

Sorry I forgot to take some fun pictures of the modules for a side by side comparison. The new Micron modules were manufactured in 2023, I think the Nemix in 2022. The surface mount components are much smaller on the Micron. The PCB traces are nearly identical but there are some distinct differences.

Some techs may be pulling their hair out at the moment reading this thread because in terms of diagnostics I essentially changed 3 variables at a time here: new UEFI/BIOS, new CPU, and new RAM.
We already know

• old UEFI/BIOS + old CPU + Nemix 128GB = Fail

However to diagnose if the old CPU was damaged we would have needed to try some other combos but I just don't have time for it now.

• old UEFI/BIOS + new CPU + old Nemix 128GB
• old UEFI/BIOS + new CPU + new Micron 128GB
• new UEFI/BIOS + new CPU + old Nemix 128GB
• new UEFI/BIOS + new CPU + new Micron 128GB (it works!)

I can take a shortcut however if that Corsair DDR4-3600 that failed with my old CPU in my ITX works in a different system without issue then I think it's fair to say my old CPU did end up taking a hit and the IMC was damaged or degraded. Right now the old CPU is paired in my ITX with the Nemix at DDR4-3200 and running perfectly stable.

 

[Ferrum Master]

Yep just JEDEC ecc.
The new Micron modules were manufactured in 2023, I think the Nemix in 2022.

Have you ever thought it is your board? Boards have additional clocking for RAM... and those clock ranges differ for each manufacturer, so there are known motherboards that needs basic capacitator/inductor changes, or termination to run certain RAM IC's. You cannot distinct from motherboard model, most probably early batches and ones fixed by hand most usually won't be identifiable, they hide those things. Have you ever thought why one board works fine with certain RAM and other not? (Yes it depends exactly on the IC properties). Do you think it is always BIOS or black magic of sort? No it is just a immature design or overcalculation. Those RAM modules are way too new and the board was designed with old AM4 design guidelines... basically a student mistake. You cannot blame AMD much here as they maintained the socket for a long time and intel always plays with RAM better, because they change sockets like old socks so they get a new redesign with fresh data.

Basically you are doing pretty adventurous task. Blaming the CPU IMC for all RAM issues ain't always right.

 

[freeagent]

I highly highly doubt you are the victim of degradation. I think you are a victim of QVL.

 

[A Computer Guy]

Have you ever thought it is your board? Boards have additional clocking for RAM... and those clock ranges differ for each manufacturer, so there are known motherboards that needs basic capacitator/inductor changes, or termination to run certain RAM IC's. You cannot distinct from motherboard model, most probably early batches and ones fixed by hand most usually won't be identifiable, they hide those things. Have you ever thought why one board works fine with certain RAM and other not? (Yes it depends exactly on the IC properties). Do you think it is always BIOS or black magic of sort? No it is just a immature design or overcalculation. Those RAM modules are way too new and the board was designed with old AM4 design guidelines... basically a student mistake. You cannot blame AMD much here as they maintained the socket for a long time and intel always plays with RAM better, because they change sockets like old socks so they get a new redesign with fresh data.

I can't completely discount it's a board compatibility issue just yet. I'll have to put the Nemix 128GB back in and see if it still has the same problems with the replacement CPU. I should have conducted this test before updating the UEFI/BIOS.

In the past ASRock fixed ECC compatibility for me before with some older Crucial (micron) DDR4-2666 modules via UEFI/BIOS update but those were manufactured in 2019. I gave them the specs of my Nemix modules including IC's and voltages but never heard back from them. There is a chance they fixed compatibility issues with the new UEFI/BIOS but that is on hopes and dreams.

Basically you are doing pretty adventurous task. Blaming the CPU IMC for all RAM issues ain't always right.

Indeed. CPU's are pretty tough and my impatience to finish with a proper test methodology with the potentially damaged chip leaves a lot of uncertainty on the table. I may have been triggered when my other system with the old (potentially damaged) CPU became unstable with my Corsair modules that were running everything stock for many months without issue.

 

[Mussels]

I think I understand what you mean but these are rated for DDR4-3200 at 1.2v.

JEDEC 3200 on my Aorus B550-I runs SoC voltage at 0.95v
XMP 3200 runs 1.09v


0.95v isnt stable with four sticks on any of my AM4 boards with 4x dual rank sticks - that's why i'm suggesting to look into those settings for stability if you mix the ram together.

Not that i suggest this, but you can make an XMP profile with the JEDEC timings, flash it to the RAM and load that identical profile and the motherboard -any brand, they all do it- will change a ton of settings in the background just because XMP is enabled.

 

[A Computer Guy]

JEDEC 3200 on my Aorus B550-I runs SoC voltage at 0.95v
XMP 3200 runs 1.09v


0.95v isnt stable with four sticks on any of my AM4 boards with 4x dual rank sticks - that's why i'm suggesting to look into those settings for stability if you mix the ram together.

I guess I won the lottery. However I probably need to do some USB load testing because SOC is related to the CPU USB ports right?

Not that i suggest this, but you can make an XMP profile with the JEDEC timings, flash it to the RAM and load that identical profile and the motherboard -any brand, they all do it- will change a ton of settings in the background just because XMP is enabled.

Ah ok.

 

[Mussels]

guess I won the lottery. However I probably need to do some USB load testing because SOC is related to the CPU USB ports right?

Everything in the SoC - it's the link between everything.
PCIE, USB, SATA, all of it. 3000 series tended to have USB issues, while 5000 tend to have PCI-E dropouts (GPU driver crashes)

 

[A Computer Guy]

So I suspect I'm finally starting to realize if my original CPU was degraded since that CPU got toasted about 17 months ago.

It's been awhile since I did the CPU swap and put that fried CPU into my gaming rig back in June 2023.

• A few months ago my gaming PC failed with various BSOD errors with my Corsair DDR4-3600 that had been running fine for around a year already prior to the CPU swap.
  • There were 3 different errors that happened in the same day
    • MEMORY_MANAGEMENT
    • IRQL_NOT_LESS_OR_EQUAL
    • Cache Hierarchy Error
  • Nothing in WHEA logs
  • I swapped in my Nemix ECC that I was previously running the CPU and things were fine
• A month ago another BSOD
  • BSOD in Memory_Management
  • Nothing in WHEA logs
  • I increased SOC to 1.1v
• Today
  • BSOD in Memory_Management
  • Nothing in WHEA logs

One thing that was interesting about the BSOD's related to Memory_Management is that the display output was really messed up in all the cases which also makes me wonder something is going on with the GPU. Usually it was like 1/2 the screen horizontally was used to display the error and squished the error to the top half of the screen and following that was some scrambled colors with the lower 3rd of the screen black.

What are the odds of two different memory kits going bad? Could this be a memory issue related to the GPU?
I still have to retest the Corsair DDR4-3600 with a different system but I really doubt anything is wrong with it.

 

[ThrashZone]

Hi,
Well gpu memory could throw the same errors I suppose as system memory does if it's timings are off :/
Graphical issues can confirm that if noticed.

 

[Mussels]

So I suspect I'm finally starting to realize if my original CPU was degraded since that CPU got toasted about 17 months ago.

It's been awhile since I did the CPU swap and put that fried CPU into my gaming rig back in June 2023.

• A few months ago my gaming PC failed with various BSOD errors with my Corsair DDR4-3600 that had been running fine for around a year already prior to the CPU swap.
  • There were 3 different errors that happened in the same day
    • MEMORY_MANAGEMENT
    • IRQL_NOT_LESS_OR_EQUAL
    • Cache Hierarchy Error
  • Nothing in WHEA logs
  • I swapped in my Nemix ECC that I was previously running the CPU and things were fine
• A month ago another BSOD
  • BSOD in Memory_Management
  • Nothing in WHEA logs
  • I increased SOC to 1.1v
• Today
  • BSOD in Memory_Management
  • Nothing in WHEA logs

One thing that was interesting about the BSOD's related to Memory_Management is that the display output was really messed up in all the cases which also makes me wonder something is going on with the GPU. Usually it was like 1/2 the screen horizontally was used to display the error and squished the error to the top half of the screen and following that was some scrambled colors with the lower 3rd of the screen black.

What are the odds of two different memory kits going bad? Could this be a memory issue related to the GPU?
I still have to retest the Corsair DDR4-3600 with a different system but I really doubt anything is wrong with it.

3600 isn't stable on every setup, it's not supported remember.
You may need as high as 1.15v SoC like I use.

Infinity fabric ties in so many components together that even PCI-E or power issues can cause these problems.

 

[A Computer Guy]

3600 isn't stable on every setup, it's not supported remember.
You may need as high as 1.15v SoC like I use.

Infinity fabric ties in so many components together that even PCI-E or power issues can cause these problems.

It seems I forgot to mention when I swapped the Corsair for the ECC after the first triple error I also dropped it down to DDR4-3200.
The subsequent 2 BSOD in Memory_Management occurred after that.

After the incidents since Dec 31 2023 I reinstalled the OS (because I was also having some other Intel wireless driver issues) and the only issue that cropped up so far was an unexplained Windows Explorer crash cycle that since has not repeated itself.

I've been distracted by my HTPC project and beta testing a game on it with the 4060LP so I haven't logged much time on my gaming rig.

 

[Mussels]

It seems I forgot to mention when I swapped the Corsair for the ECC after the first triple error I also dropped it down to DDR4-3200.
The subsequent 2 BSOD in Memory_Management occurred after that.

After the incidents since Dec 31 2023 I reinstalled the OS (because I was also having some other Intel wireless driver issues) and the only issue that cropped up so far was an unexplained Windows Explorer crash cycle that since has not repeated itself.

I've been distracted by my HTPC project and beta testing a game on it with the 4060LP so I haven't logged much time on my gaming rig.

My x370 system would have cold boot issues at 3200 but was fine at 3133 - sometimes it's just tuning the bastards in.

Ah yeah that's right, the JEDEC RAM.

That'll probably make the auto settings be weird as heck, since there really is assumptions about XMP being used at higher ram speeds. I've got no idea what settings might be left unchanged with that.

 

[A Computer Guy]

My x370 system would have cold boot issues at 3200 but was fine at 3133 - sometimes it's just tuning the bastards in.

Ah yeah that's right, the JEDEC RAM.

That'll probably make the auto settings be weird as heck, since there really is assumptions about XMP being used at higher ram speeds. I've got no idea what settings might be left unchanged with that.

So memory wise I haven't had any further issue after setting the SOC to 1.1v and LLC to some mid level setting. In prior UEFI/BIOS revisions than what I have now I remembered LLC being in the middle but after the last update it ended up getting reset to it's lowest LLC value.

 

[A Computer Guy]

Continuing updates... (at 19 months since CPU was toasted)

Today on my gaming rig I was having a verity of errors again starting with Ryzen Master not displaying telemetry data along with MEMORY_MANAGEMENT and Cache Hierarchy Error errors. Along with this problem recently I was encountering a rapid increase in USB-C switch failure when the gaming PC was connected rendering me unable to use the switch to change keyboard and mouse between PC's until it was disconnected from the Gaming PC.

So I decided to swap the 5950x for the 5800x (that I got on sale awhile back) to see if I can isolate if the 5950x (the one that was originally cooked) is a problem or not. Unfortunately I still encountered WHEA uncorrectable error and Cache Hierarchy Error after the swap so then I decided to swap out my 64GB Nemix ECC kit too for the other one I had on the shelf since I couldn't use them combined as 128GB in the Taichi.

Things seem to be working ok again. I could say oh well it seems the first Nemix kit was the problem but that doesn't explain why the failures started occurring with my Corsair kit originally (also the recent USBC problem) so there is a possibility my ITX motherboard might actually be part of the problem. :-(

At stock the 5800x seems to have a decent score with NH-C14S with CB-23 reaching steady state about 82c @4.46Ghz.
I'll have to keep an eye on this for awhile and see how it works out.