Overclock.net banner
1 - 20 of 105 Posts

·
Registered
Joined
·
73 Posts
Discussion Starter · #1 ·
AMD Ryzen 5000 (Zen3) Whea 19 Discussion Thread


Please remember the following


Whenever posting about your experiences please provide information about the system, specificly of interest should be:

CPU Type, Memory IC, Mainboard, UEFI / Bios / Agesa version and voltages used


This thread is specificly for Whea 19 Warnings not for 18s or 20s Errors, which should be related to instability.
Try to stay on topic and keep it constructive



I will keep summing up our Findings here, when we reaced a consensus:


Happy posting!​
 

·
Registered
Joined
·
73 Posts
Discussion Starter · #2 ·
My Initial Findings:
  • I was able to fix my USB Dropouts and Sound Issues by increasing the vddg iod voltage, I think those are not related to the wheas at all
  • Memory topology does not seem to matter (my B550 Vision D has one of the strongest topologies out there and I still had whea 19s @ 1866MHz fclk on some Agesa Versions)
  • Agesa / Uefi / Bios versions matter a LOT for fclk (1.2.0.2 and 1.2.0.3 Patch A fixed my fclk hole and allowed me to reach fclks over 2000, 1.2.0.3 Patch B introduced wheas as low as 1866 which was running fine on older Bios revisions).
  • Realtek Lan chips don't seem to be the issue (my board does not have one) but the Realtek Driver included in the Agesa could cause issues
  • Single CCD Chips seem to have an easier time with higher fclks and the wheas
  • Seems unrelated to timings as long as your settings are stable
  • Seems unrelated to 4x8 / 2x16 / Daisy Chain / Topology
  • Some combinations of vsoc, vdimm,vddg, vddg iod, vddg ccd & vddp can help to have less whea 19s, but can not get you whea 19 free.

Speculation on my part from reports I read:
  • B-Dies seem to have more issues with Whea 19s than ICs that are "easier" on the memory controller
  • Board Size / PCIe Topology seems to matter
  • Buildzoid reported in one of his livestreams that Mann1x's Whea Supressor does not fix the performance issues correlating with the whea 19s

or maybe it's just broken (I remember Zen3 having issues with PCIe 4.0 instability on launch)
 

·
Premium Member
Joined
·
20,738 Posts
So if I settle for 3600 Cl16 I won't have to deal with this Whea 19?

Auros B550M Pro and 5900X. Windows 1909 and 20H2.

Never had USB dropouts on this board having an R5 3600 for about a year and about 4 mos with a 5900X.
 
  • Rep+
Reactions: rossi594

·
Registered
Joined
·
812 Posts
5950x here, don't have any FCLK holes that I can find, have b-die rated for 4000 14-15-15-15, when running it on XMP/auto and FCLK 2000 I have found it hard to stop the USB issues.

Apart from the WHEA spam everything appears to be running OK, some shorter TM5 runs pass OK, but the quickest way for me to produce mass USB dropouts was to run y-cruncher test 17, Hybrid NTT. USB would go crazy within 30 seconds, such as my Corsair commander pro going mad. I would warn anyone that uses a commander pro to handle their watercooling/fan curves that not only will this nuke your custom profiles, it made my device think it even needed a FW update to function.

So I would say to anyone trying anything above 1900 WHEA run y-cruncher test 17 to really see if you are stable/have sorted USB issues. I tried IOD up to 1.15v.
 

·
Registered
Joined
·
73 Posts
Discussion Starter · #5 ·
So if I settle for 3600 Cl16 I won't have to deal with this Whea 19?

Auros B550M Pro and 5900X. Windows 1909 and 20H2.

Never had USB dropouts on this board having an R5 3600 for about a year and about 4 mos with a 5900X.
I've heared of noone having whea 19 warnings at 3600. You should be fine.

5950x here, don't have any FCLK holes that I can find, have b-die rated for 4000 14-15-15-15, when running it on XMP/auto and FCLK 2000 I have found it hard to stop the USB issues.

Apart from the WHEA spam everything appears to be running OK, some shorter TM5 runs pass OK, but the quickest way for me to produce mass USB dropouts was to run y-cruncher test 17, Hybrid NTT. USB would go crazy within 30 seconds, such as my Corsair commander pro going mad. I would warn anyone that uses a commander pro to handle their watercooling/fan curves that not only will this nuke your custom profiles, it made my device think it even needed a FW update to function.

So I would say to anyone trying anything above 1900 WHEA run y-cruncher test 17 to really see if you are stable/have sorted USB issues. I tried IOD up to 1.15v.
Interesting, did you run vsoc high enough for the power to be there to supply the vddg iod? Maybe try some more vddg ccd, the problem could be on the other end of the infinity fabric.
 

·
Registered
Joined
·
812 Posts
I've heared of noone having whea 19 warnings at 3600. You should be fine.



Interesting, did you run vsoc high enough for the power to be there to supply the vddg iod? Maybe try some more vddg ccd, the problem could be on the other end of the infinity fabric.
VSOC was 1.2v, CCD was 1.05v. I got frustrated with it nuking my commander pro settings continually and also worried it could be on the verge of nuking my Windows installation lol.

Next time around I was going to try going higher than 2000, because it boots fine and there was some advice for whatever reason stabilising higher might be "easier".

Given that I've been able to run IOD and CCD as low as 0.925v at 3800/1900 with zero WHEA, it just got me thinking is this really worth it for such high voltage increases lol. Though right now I'm at 0.975v CCD and IOD 1.05v due to scaling and/or seeing good performance. No real reason to run voltages that low if I can get better performance higher.

But VSOC above 1.2v and VDDGs reaching 1.15v+ just seems excessive. Hopefully AMD can get their head pulled out their arses and do more work on FCLK above 1900.
 

·
Registered
Joined
·
739 Posts
 

·
Registered
Joined
·
73 Posts
Discussion Starter · #9 ·
Sorry what do you want to tell us?
 

·
Overclock the World
Joined
·
3,282 Posts
or maybe it's just broken (I remember Zen3 having issues with PCIe 4.0 instability on launch)
PCIe dropouts @ 2000 FCLK are known to AMD - hence the Lanes cut on Cezanne U
Matisse could run 2000 FCLK, if it would be allowed to do so. People got it up to 1966 till AMD ~ ABL Locked it to 1900 , to make PCIe 4.0 look better & more stable

Vermeer can run 2000 FCLK, i stand by my word ~ every unit / when they fix the error reporting sensorics (if they even want to)
Higher is dependent on too many things. 2033 the community here should be capable in doing (knowledge based). 2000 runs with XMP on 1:1 perfectly fine

SOC & FCLK voltage predictions are set up till 2067 quite "fine" , but Matisse people need to get more comfortable running beyond 1.2v SOC.
Auto for 2033 & 2067 is 1.25v , till hard peak 1.3v. Do not exceed 1.3v, on any scenario :)
Buildzoid reported in one of his livestreams that Mann1x's Whea Supressor does not fix the performance issues correlating with the whea 19s
We talked with Buildzoid slightly. Surpressor only is there to fix DPC spam call
It can not fix the user not having to optimize 1.8VDD rail , VDDG IOD & SOC.

#19 is not a memOC related issue, to be shown as "lower performance with CPU Benchmarks"
It's neither a CPU nor IMC issue.
It has a connection to the PCH (I/O issue) but is also not that ~ soo it clearly is not a PCIe 4.0 issue
It's a bad sensorics bug issue ~ a DPM/GMI Link power management issue

High FCLK will push higher SOC
CPU on stock limits was designed to perform without throttle @ 3600mhz, with maybe passing on 3800 if undervolted
Anything slightly higher, will cut deep into the powerbudget and lower boost ~ without having anything to do with the CPU/Boost System itself
This has to be hopefully now known ~ that at any memOC , people have to extend the artificial stock limits slightly ~ else Cache & Boost throttles
 

·
Registered
Joined
·
73 Posts
Discussion Starter · #11 ·
PCIe dropouts @ 2000 FCLK are known to AMD - hence the Lanes cut on Cezanne U
Matisse could run 2000 FCLK, if it would be allowed to do so. People got it up to 1966 till AMD ~ ABL Locked it to 1900 , to make PCIe 4.0 look better & more stable

Vermeer can run 2000 FCLK, i stand by my word ~ every unit / when they fix the error reporting sensorics (if they even want to)
Higher is dependent on too many things. 2033 the community here should be capable in doing (knowledge based). 2000 runs with XMP on 1:1 perfectly fine

SOC & FCLK voltage predictions are set up till 2067 quite "fine" , but Matisse people need to get more comfortable running beyond 1.2v SOC.
Auto for 2033 & 2067 is 1.25v , till hard peak 1.3v. Do not exceed 1.3v, on any scenario :)

We talked with Buildzoid slightly. Surpressor only is there to fix DPC spam call
It can not fix the user not having to optimize 1.8VDD rail , VDDG IOD & SOC.

#19 is not a memOC related issue, to be shown as "lower performance with CPU Benchmarks"
It's neither a CPU nor IMC issue.
It has a connection to the PCH (I/O issue) but is also not that ~ soo it clearly is not a PCIe 4.0 issue
It's a bad sensorics bug issue ~ a DPM/GMI Link power management issue

High FCLK will push higher SOC
CPU on stock limits was designed to perform without throttle @ 3600mhz, with maybe passing on 3800 if undervolted
Anything slightly higher, will cut deep into the powerbudget and lower boost ~ without having anything to do with the CPU/Boost System itself
This has to be hopefully now known ~ that at any memOC , people have to extend the artificial stock limits slightly ~ else Cache & Boost throttles
So he just didn't tune his voltages? As always you've been very insightfull, thanks for posting.
 

·
Overclocked / Overvolted
Joined
·
237 Posts
After giving it a great deal of thought and hearing all of the fanboys chirping about how great Zen 3/Ryzen 9 CPUs are, I took a leap of faith and really regret it. AMD still sucks as much as they did a decade ago when I swore I would never spend another dime on anything with their name on it. I should have kept that promise to myself. I sold a nicely binned 10900K that I benched as high as 5.5GHz on chilled water and a Z490 Dark K|NGP|N mobo for this and I can say that is was a very bad decision that has left me disappointed with the outcome.

I can tell you my experience has been that the USB dropout problem is more frequent and more annoying the harder I push the CPU and memory controller. If I use sissy-boy PBO overclocking and run my memory at mediocre clock speeds with sloppy timings the problem is not nearly as frequent and aggravating. I bought the CPU and mobo because my passion is overclocked benching. If I partially mitigate the USB drop-out problems by running a mundane configuration, the system immediately becomes unfit and worthless for the purpose for which it was purchased.

The specs on the system are in my signature. I have tried every BIOS released that supports 5950X from v2502 forward and don't notice much change. I am currently on BIOS v3801 and tried the 3901 beta. AGESA V2 PI 1.2.0.3 Patch C. I am running CSM with Legacy only, TPM options disabled, virtualization and no-execute security option disabled.

I have 3 different memory kits (all Samsung B-die). 2*8GB Single Rank Patriot Viper DDR4-4000CL17, G.SKILL Ripjaws V in 2*8GB DDR4-4000CL15 and 2*16GB Dual Rank DDR4-4000CL17. Some overclock more gracefully than others, but the USB drop-out issue is the same no matter which modules are installed. I have no WHEA errors at 1900 FCLK or lower. They start at 2000, so I can't run 1:1 without issues unless I stay at 3800 or below.

I am running CCD0 at 47x and CCD1 at 46x for my daily driver overclock. I have VID set at 1.500V and manual core voltage at 1.375V. No limits on watts or amps. I have SOC at 1.295-1.300V. cLDO VDDP, VDDG IOD and CCD are all set to 1.0979V. I have DIMM voltage set at 1.500V running the dual rant 4000 sticks at 4000 CL 16-17-16-36 2T and custom timings. Using XMP and more conservative memory clock speeds, lower FCLK to 1900 or less and lame timings, higher and lower voltage on the RAM provides no relief for the USB drop-out problem. It does mitigate the WHEA CPU Bus Interconnect issues, but it sucks to have to run a mundane configuration to avoid the WHEA errors and make the system totally worthless to me.

Even if this eventually gets fixed, this has caused my hate for AMD to return and it is unlikely I will ever purchase anything from them again. I should have stuck with what I knew produced consistently acceptable results (Intel). I was hopeful a decade was long enough to hold a grudge for the garbage they produce, but apparently not.

I heard rumors that disabling c-states and manually setting Gen 3 PCIe 3.0 helps with the USB drop-outs. I haven't found that to be true with c-states. I normally disable that on all systems anyway. It was already disabled as part of my normal BIOS configuration and it seems to have no effect on this problem. I have not manually set the Gen 3 for PCIe. Maybe I should. It sucks to have to turn off something that is one of the overhyped features of the platform, but it doesn't really provide any benefit in the grand scheme of things. I will try that and report if it seems to help.
 
  • Rep+
Reactions: rossi594 and 0451

·
Overclock the World
Joined
·
3,282 Posts
Never had USB dropouts on this board having an R5 3600 for about a year and about 4 months with a 5900X.
Gladly i never had them too
But ironically they start to come with the experimental 1.2.0.0 patches and onward (which should fix it, not introduce them)
1203 now fixed things - and they are gone again
But it still is connected to proper VDDG voltages (low CCD and high IOD) :)

What one should disable are DF_C-States, not global C_States
C6 states are buggy and cause overboost bugs & overvoltage bugs
Soo this includes Powersuply idle control - to typical , instead of low current (on some boards)
 

·
Overclocked / Overvolted
Joined
·
237 Posts
Gladly i never had them too
But ironically they start to come with the experimental 1.2.0.0 patches and onward (which should fix it, not introduce them)
1203 now fixed things - and they are gone again
But it still is connected to proper VDDG voltages (low CCD and high IOD) :)

What one should disable are DF_C-States, not global C_States
C6 states are buggy and cause overboost bugs & overvoltage bugs
Soo this includes Powersuply idle control - to typical , instead of low current (on some boards)
What motherboard and CPU are you running? I would have global c-states disabled in any case. I will look for DF_C-States and disable it when I find it. The helter-skelter menus in ASUS's AMD BIOS environment is very disorganized and confusing, riddled with redundancy and menu options that don't actually change anything. It is very disappointing what a sloppy, half-assed outfit they have become. Firmware used to be their primary area of strength and excellence.

I already have the Power Supply Idle control set to typical. I generally disable all forms of power management in the BIOS and in Windows.
 
  • Rep+
Reactions: rossi594

·
Overclock the World
Joined
·
3,282 Posts
I already have the Power Supply Idle control set to typical. I generally disable all forms of power management in the BIOS and in Windows.
Ryzen without any powermanagement, does eat too much into the powerbudget
Ryzen on an allcore, does degrade the unit (if overdone) and performs slower . Working against FIT is not a good idea

Specs are in the Signature. Can not complain about AMD FW ~ what has to work, works & ASUS's bioses usually show a lot of AMD CBS ~ which has to be that way but is sadly a rarity
Except for C6 states, who waste useless power & lowers boost that way
 
  • Rep+
Reactions: rossi594

·
Overclocked / Overvolted
Joined
·
237 Posts
Ryzen without any powermanagement, does eat too much into the powerbudget
Ryzen on an allcore, does degrade the unit (if overdone) and performs slower . Working against FIT is not a good idea

Specs are in the Signature. Can not complain about AMD FW ~ what has to work, works & ASUS's bioses usually show a lot of AMD CBS ~ which has to be that way but is sadly a rarity
Except for C6 states, who waste useless power & lowers boost that way
Thanks. I did not realize your signature block was collapsible until you said that your specs were there.

I am not interested in conserving energy, I only care about overclocking my hardware to the edge of functionality because that's my hobby and that is the only reason I purchased the 5950X and Crosshair VIII motherboard. I don't use boost, only static core ratios and static voltage and set all power limits to the maximum allowed. Boost does not allow all cores to clock as high as I clock them manually, and my benchmark scores are lower when using PBO overclocking. I am not worried about the longevity of the CPU. That concern is not compatible with my hobby. If the CPU degrades I will throw it away and buy another one, but I am probably not going to waste my money on anything from AMD ever again. I use the term "probably" and won't say never because I do make exceptions. I made one for the 5950X, and it turns out that it was poor judgment on my part.
 
  • Rep+
Reactions: rossi594

·
Overclock the World
Joined
·
3,282 Posts
I don't use boost, only static core ratios and static voltage and set all power limits to the maximum allowed. Boost does not allow all cores to clock as high as I clock them manually, and my benchmark scores are lower when using PBO overclocking.
Soo you exceed 5.1Ghz allcore with fixed voltage ?
Else, per core curve optimizer is the way to go - or investing into Hydra by 1usmus ~ as that runs under OC_Mode & can lead you to 5.2~5.3ish up to loadtype
Balancing 16 cores is not an easy task, a fixed voltage is easier ~ but performance will be worse

Hard to explain, we are used to different methodic's of overclocking
In general, fixed clock ~ the last unit was Pinnacle Ridge (2700X) where it worked
Since then, Precision Boost ~ same as on Intel Skew Voltage & Boost Skew control, is the "alternative / well current" way of overclocking them
Core suspension is needed not only for powerdraw, but also for in-chip signal integrity.
Same as Dynamic FCLK (if AMD ever brings that thing back) will be much better than fixed fabric frequency.

Long topic , and it's impossible to agree on something together. We took different approaches :)
Yet, please don't forget that there are IPC differences, and not only clock number differences
I look forward to both releases Zen3D and Alder Lake. Both deliver promising results

EDIT:
Speaking about pushing
Once you have only WHEA #19, run the WHEA suppressor
keep in mind, FIT will also push it's hands even on OC_Mode (allcore mode)
Soo you have to extend the powerlimits and other cTDP, Package Power limits.

Voltage control by the VRM is not detected - offsets of it
Vermeer is very capable to boost beyond 1.65v at near 6ghz ~ if you find a method to bypass FIT
While i absolutely do not recommend it ~ keep in mind, FIT doesn't detect core offsets & V/F Curve extends far into the 5.8-6ghz range. It has steppings, just needs the voltage (VID) to try and run it
Happy XOC :)
 

·
Premium Member
Joined
·
20,738 Posts
Gladly i never had them too
But ironically they start to come with the experimental 1.2.0.0 patches and onward (which should fix it, not introduce them)
1203 now fixed things - and they are gone again
But it still is connected to proper VDDG voltages (low CCD and high IOD) :)

What one should disable are DF_C-States, not global C_States
C6 states are buggy and cause overboost bugs & overvoltage bugs
Soo this includes Powersuply idle control - to typical , instead of low current (on some boards)
You must have an Asus motherboard. I have only used one BIOS version on my Gigabyte so far with the 5900X and i plan to stick to it since it works. But i was using this B550M Pro for about a year with an R5 3600 and that too never had USB issues with the original BIOS - F1. Jumped to F13 prior to switching cpu. I have a B350F, an X470 Prime Pro, and an A320 and never had any USB issues with those too. Just lucky i guess.

I don't use PBO. Tried it last week but did not like it. Just a slight oc using BCLK is all that is needed.
 
  • Rep+
Reactions: Veii

·
Overclocked / Overvolted
Joined
·
237 Posts
Soo you exceed 5.1Ghz allcore with fixed voltage ?
Else, per core curve optimizer is the way to go - or investing into Hydra by 1usmus ~ as that runs under OC_Mode & can lead you to 5.2~5.3ish up to loadtype
Balancing 16 cores is not an easy task, a fixed voltage is easier ~ but performance will be worse

Hard to explain, we are used to different methodic's of overclocking
In general, fixed clock ~ the last unit was Pinnacle Ridge (2700X) where it worked
Since then, Precision Boost ~ same as on Intel Skew Voltage & Boost Skew control, is the "alternative / well current" way of overclocking them
Core suspension is needed not only for powerdraw, but also for in-chip signal integrity.
Same as Dynamic FCLK (if AMD ever brings that thing back) will be much better than fixed fabric frequency.

Long topic , and it's impossible to agree on something together. We took different approaches :)
Yet, please don't forget that there are IPC differences, and not only clock number differences
I look forward to both releases Zen3D and Alder Lake. Both deliver promising results

EDIT:
Speaking about pushing
Once you have only WHEA #19, run the WHEA suppressor
keep in mind, FIT will also push it's hands even on OC_Mode (allcore mode)
Soo you have to extend the powerlimits and other cTDP, Package Power limits.

Voltage control by the VRM is not detected - offsets of it
Vermeer is very capable to boost beyond 1.65v at near 6ghz ~ if you find a method to bypass FIT
While i absolutely do not recommend it ~ keep in mind, FIT doesn't detect core offsets & V/F Curve extends far into the 5.8-6ghz range. It has steppings, just needs the voltage (VID) to try and run it
Happy XOC :)
Maybe my CPU is a poor silicon sample. That might explain my disappointment with Ryzen. But, PBO and curve optimizer overclocking simply do not work worth a damn for me. About the most I can manage to squeeze out of it is about 4.4-4.5GHz max under Cinebench and the scores are a couple hundred points less than manual all-core overclocking using fixed voltage. I support 1usmus on Patreon and have Hydra. It works well and he has done a great job on it, but I still get better benchmark scores doing it the old fashioned way. Using Hydra do produce better results for me than trying to rely on the BIOS options that do not work well for me. I do not do sub-zero overclocking, but I do my overclocked benching on chilled water. Without the chilled water, it runs far too hot to be impressive inn terms of overclocking capacity. I am going to delid it and see if it helps as much as it always had on Intel. Even all of my Intel CPUs with a soldered IHS benefitted from delidding, so I am hoping my 5950X will as well. I wish I could find a bare die solution for the 5950X. That would be ideal. My last two Intel CPUs were run bare die and everything just works so much better that way.
 
  • Rep+
Reactions: rossi594

·
Robotic Chemist
Joined
·
4,302 Posts
I am going to delid it and see if it helps as much as it always had on Intel. Even all of my Intel CPUs with a soldered IHS benefitted from delidding, so I am hoping my 5950X will as well. I wish I could find a bare die solution for the 5950X. That would be ideal. My last two Intel CPUs were run bare die and everything just works so much better that way.
All the examples I have seen of a delidded 5950X were not good (I haven't seen many). The issue is that the chiplets are not at exactly the same height, so the advantage from delidding (very thin thermal interface material) is not there. At best the same as stock. :(
 
1 - 20 of 105 Posts
Top