5950X started giving out Whea Buserrors. Looking for OC that would be more stable and what safe voltages to use.

[EmperorOfFinland]

So basically im sorta stuck here, i needed to bring the clocks back on the ram to XMP because of unstability after 3-5h of rendering with the cpu.
I have 2x 16GB sticks on the pc

Now inorder to get them back i would need data on the safe voltages on the cpu and the fabric, i run the cpu as stock as i dont see point on overclocking as the 360mm radiator is not enough to keep the 5950x under 80-90C
As the computer is not fully stable with the settings, it would not crash out on normal usage and the bus errors were so rare that i would get them once per month. and the only way of actually crashing the pc is to render 3-6h of my youtubevideos as a memorytest does not bring out the errors nowadays.

So this is the ram i have on my pc and i was thinking i would stuff this to the ram as the ram is not that good to begin with just to get some performance and get the pc not bluescreen when overclocked.

Albeit if you have better settings to try on the machine as its the first time i have been doing memory overclocking myself so i have no idea where this memory would go to or where to take it since ryzen 5000 settings are not on the calculator.

 

[Toothless]

Usually stock for 3000 and 5000 are the way to go since they technically overclock themselves. As far as we know on 7nm you do not want to pass 1.28v on manual overclocks, full load. I don't know if that's completely the same for 5000.

 

[EmperorOfFinland]

Yeah i have figured the same thing out. the overclock wont matter for gaming atleast now the 1-3fps is not really worth bothering to OC as i degraded one ryzen 3800x with overcloking so im not intending to do that as no one really knows the safe voltages for anything.

 

[phanbuey]

if it is at stock settings and under warranty then send it back.

 

[Toothless]

Yeah i have figured the same thing out. the overclock wont matter for gaming atleast now the 1-3fps is not really worth bothering to OC as i degraded one ryzen 3800x with overcloking so im not intending to do that as no one really knows the safe voltages for anything.

Now, was that with too much voltage? I ask because we have a few people here who think 3xxx can take 1.4v because Pilldriver/Bulldozer did..

 

[EmperorOfFinland]

its not at stocksettings i had it at 3800mhz CL16, but im planning to put it back there as there is a big latency gap. now the kit has worked before and i think the issue is infinityfabric related as i have brought the kit to 3733mhz and 3600mhz at tighter latencies with the same kit. problem is that. I have had the kit since 2019 december and it has worked without errors, i brought it back to xmp since i dont have working overclocked settings for it at 3800mhz and i know the memory would work easily at thouse lower settings and i cant have the pc be crashing when i do my youtuberenders as they take 1-2h per render.

Also the voltage was the reason. so the chip lost 100mhz of boostclocks i did some experiments with cinebench at 1.45v and noted that 1.35v is not even safe. so i returned it back to stock. on old intels people about know what is the tolerable voltage but on ryzen no one knows. they are just plainly guesses. The chip is still alive on my brothers rig and doing just fine so the chip never died. Based on my experiences stuffing above 1.3v seems like extremely bad idea.

 

[Toothless]

its not at stocksettings i had it at 3800mhz CL16, but im planning to put it back there as there is a big latency gap. now the kit has worked before and i think the issue is infinityfabric related as i have brought the kit to 3733mhz and 3600mhz at tighter latencies with the same kit. problem is that. I have had the kit since 2019 december and it has worked without errors, i brought it back to xmp since i dont have working overclocked settings for it and i know the memory would work easily at thouse settings and i cant have the pc be crashing when i do my youtuberenders as they take 1-2h per render.

Also the voltage was the reason. so the chip lost 100mhz of boostclocks i did some experiments with cinebench at 1.45v and noted that 1.35v is not even safe. so i returned it back to stock. on old intels people about know what is the tolerable voltage but on ryzen no one knows. they are just plainly guesses. The chip is still alive on my brothers rig and doing just fine so the chip never died. Based on my experiences stuffing above 1.3v seems like extremely bad idea.

The theory is 1.275v, but you really don't wanna go above 1.3v. I'm at 1.287v or so for mine and it's good so far. Thank you for putting that in!

 

[EmperorOfFinland]

So i was merely of asking for better 3800mhz memory OC settings i could try around over the weekend to see what is stable. it does not have to be super fast but i literally have 0 idea about the voltages relating to ryzen 5000 on the infinityfabric and memory itself. i will run CPU as stock as there is no point of even bothering to oc it. Working with the settings on the 2nd picture that i posted on the first post. Something similar but actually fully stable as the computer does not show errors on memtest and on other things, it just crashes out after 3-5 hours of rendering so its pain in the ass to test does it crash. also the errors are so rare that i might get 1-2 of them per month eventhough i play games and render on the machine so there might be 1-2 months before the computer crashes again.

I updated to the newest bios on Asus Prime-P X570 to fix the usb dropout with the usb expansioncard i had and now it has been fully working after the biosupdate but i think the bios made the system more unstable since they had to fix that.

 

[Toothless]

Was it an AGESA update? AMD bios updates are usually pretty good in terms of stability improvements

 

[EmperorOfFinland]

Yeah it was agesa update and fix for the USB dropout issue i was having with the card. the known issue

 

[GerKNG]

AMD really needs an "Pre Alpha" Badge on their CPU boxes..

 

[EmperorOfFinland]

Its probably just the infinityfabric being what it is and nothing too severe. i know the memory can do 3800mhz at thouse timings but seems the fabric cant. so option here is either drop frequency and tighten the ram. im sure if i dropped the frequency to 3733mhz it would run without problems. or try revised 3800mhz settings with revised voltages to see can i get it stable

 

[Zach_01]

Hi there...

What is your current configuration of DRAM? Other than 3800MHz.
Use latest ZenTimings software.

The RyzenDRAMcalculator may work better if you import XMP profile of your RAM on it.
You can save a report of XMP with Thaiphoon Burner.
Open it and click the "Report" next to "Read". Scroll all way down and click the "Show delays in clock cycles" once to turn into "Show delays in nanoseconds".
Then click "Export" and "Complete HTML Report".
This file you can import in calculator with the "Import XMP" button in main page. Then you click the safe/fast presets.

Ryzen 5000 have the cool feature of CurveOptimizer build in PBO settings. OC is very easy with it, but remember to not over do it. Things can get easily out of hand and power consumption (CPU PPT) and silicon stress under high current(amperage/EDC) can skyrocket. Low temp is the key for all Ryzens. Under 80C if it’s possible for max sustained. 90C is max operating temp for 5000.

Use HWiNFO64 (sensor mode only) for a complete operating report of CPU and system. There you can monitor many many things and PPT/EDC are among them.

There is no straight answer for the safest voltage on a ZEN2/3 CPU. Depends on the temperature + use case scenario(type of load) = Current (EDC).

Give ZenTimings and HWiNFO (sensor mode) a look and feel free to ask anything.

Have you check on Windows Event Viewer the type of WHEA errors you’re getting?

Also IMHO when someone is DRAM OCing it’s better to disable all power savings for DRAM and SoC/IF in BIOS.
And there are a few reports with Ryzen5000 giving WHEA errors on random idle conditions when the system is resting or doing almost nothing. So my first thoughts on it is to try to disable power savings.

PowerDownMode: Disabled
DF C-States: Disabled
SoC/Uncore OC Mode: Enabled

PowerDown is for DRAM. DF is DataFabric (=InfinityFabric). SoC/Uncore is of course the SoC/IO Die of the CPU containing the memory controller (UMC/UCLK) among others.

 

[Deleted member 205776]

AMD really needs an "Pre Alpha" Badge on their CPU boxes..

Thank you for your useful contribution to the discussion. I think board manufacturers need a "Stop Overvolting" sticker plastered onto their hand.

What I did to stabilize my WHEAs (by stabilize I mean removing them completely) was fiddling around with the SoC voltage and its derivatives, namely lowering them. My board overvolted my SoC, VDDG and VDDP voltages way too much for the RAM kit that I have (1.1 on SoC, 1.05 on both VDDGs and 0.900 on VDDP, for 4 sticks of 3600 MHz). Caused idle reboots and random WHEAs. They're all gone now.

What first fixed them was lowering VDDG IOD and leaving everything else stock. But then I decided to just lower all of them and it was still stable.

The overvolting wasn't as bad on my ROG Strix B550-F board compared to my Gigabyte B550 Aorus Pro board. The Gigabyte board decided to give 1.1 on the SoC, 1.1 on BOTH VDDGs and 1.0v on VDDP. For what kit you may ask? Well a 3200 MHz B-Die kit of course. Completely dumb and overkill. That caused reboots and WHEAs within 2 minutes of being in Windows. I didn't know at the time so I just disabled C-states and it worked. The same board proceeded to die days later, with the replacement being DOA.

But now that I know, I have no issues keeping C-states on and my power supply on Low Current Idle, because I fixed the voltages.

 

[Durhamranger]

You may want to check out this thread on overclock.net https://www.overclock.net/threads/w...nfinity-fabric-overclocking-headroom.1778117/

Anything newer than windows 10 version 1909 is WHEA bug fest........

 

[Ferrum Master]

None of the screenshots show actual timings and voltages.

The Whea errors contain information what core crashed also. Seek it out if it is random or constant.

I would not touch any tool from 1usmus when using Ryzen 5000.

 

[Durhamranger]

Yup best of posting a ZenTimings Screenshot like so...

 

[EmperorOfFinland]

Yeah i have been running the 2000 version windows for a while i was running the exact settings i posted on the screenshot before on the ramtuner. and thinking of revising them. This is what im running currently as its XMP and the computer does not crash under that or give errors as im not pushing the system at all. But i wonder what settings i should use for that. So basically what the ramtuner screenshot shows on the first one is what settings i used for the 3800mhz overclock

 

[Calmmo]

Not running 1:1 based on that shot

 

[GerKNG]

Thank you for your useful contribution to the discussion. I think board manufacturers need a "Stop Overvolting" sticker plastered onto their hand.

What I did to stabilize my WHEAs (by stabilize I mean removing them completely) was fiddling around with the SoC voltage and its derivatives, namely lowering them. My board overvolted my SoC, VDDG and VDDP voltages way too much for the RAM kit that I have (1.1 on SoC, 1.05 on both VDDGs and 0.900 on VDDP, for 4 sticks of 3600 MHz). Caused idle reboots and random WHEAs. They're all gone now.

What first fixed them was lowering VDDG IOD and leaving everything else stock. But then I decided to just lower all of them and it was still stable.

The overvolting wasn't as bad on my ROG Strix B550-F board compared to my Gigabyte B550 Aorus Pro board. The Gigabyte board decided to give 1.1 on the SoC, 1.1 on BOTH VDDGs and 1.0v on VDDP. For what kit you may ask? Well a 3200 MHz B-Die kit of course. Completely dumb and overkill. That caused reboots and WHEAs within 2 minutes of being in Windows. I didn't know at the time so I just disabled C-states and it worked. The same board proceeded to die days later, with the replacement being DOA.

But now that I know, I have no issues keeping C-states on and my power supply on Low Current Idle, because I fixed the voltages.

the stock core voltage is completely governed by the CPU and the motherboard has no influence on it (stock!)

thanks for your tons of tweaks to get a CPU barely working at stock out of the box.
that's totally how it's supposed to work! ...

like i said. pre alpha.

 

[Zach_01]

Yeah i have been running the 2000 version windows for a while i was running the exact settings i posted on the screenshot before on the ramtuner. and thinking of revising them. This is what im running currently as its XMP and the computer does not crash under that or give errors as im not pushing the system at all. But i wonder what settings i should use for that. So basically what the ramtuner screenshot shows on the first one is what settings i used for the 3800mhz overclock

View attachment 197363

MCLK, FCLK and UCLK "must" run at same speed for best performance. Otherwise you got a latency penalty and same goes for memory performance. Memory controller speed (UCLK) has synched at half the memory speed (MCLK).
This is typical behavior when InfinityFabric (FCLK) is out of synch with MCLK. For some reason FCLK is set to 1600MHz instead of 1800MHz, or you set it there by accident(?).

It should look like this. All at same speed at whatever speed you choose

 

[Ferrum Master]

It is a mess there for sure. I have spent few weeks finding what really works with my RAM. I have old set of Micron D-die. There is almost zero information about it.

Basically you have to slowly tinker and adapt the settings one by one. I do not have any WHEA errors, Linpack stable for 24h. I raised few more voltages also, but only a bit, because I drive 4 modules, 2 may not need it. I can boot my ram at 3600 even. But actually have diminishing returns. I have to run very high tRFC to be stable and it steals all the performance boost, it is the old Micron IC limitation. The only solution for low tRFC are Samsung B dies. There's so much wonky stuff in that screenshot, it is hard to note where to start.

 

[EmperorOfFinland]

Oh right so it seems. let me fix that tomorrow on the bios. stupid XMP doing its own stuff and me selecting the wrong setting myself

Done i fixed the infinityfabric clocks. i just need to figure out how to push thease.
i can boot the ram at CL16 3800mhz if needed but infinityfabric does not seem stable. its hynix C/D die ram

 

[thesmokingman]

Oh right so it seems. let me fix that tomorrow on the bios. stupid XMP doing its own stuff and me selecting the wrong setting myself

Done i fixed the infinityfabric clocks. i just need to figure out how to push thease.
i can boot the ram at CL16 3800mhz if needed but infinityfabric does not seem stable. its hynix C/D die ram
View attachment 197572

Why even bother oc'ing that ram?? You didn't buy very fast ram to begin with so I'm really scratching my head, why bother? The long render sessions taxes the ram and memory system so eeking out a couple nanoseconds that no human will ever notice... what is the point?

 

[Makaveli]

So basically im sorta stuck here, i needed to bring the clocks back on the ram to XMP because of unstability after 3-5h of rendering with the cpu.
I have 2x 16GB sticks on the pc

Now inorder to get them back i would need data on the safe voltages on the cpu and the fabric, i run the cpu as stock as i dont see point on overclocking as the 360mm radiator is not enough to keep the 5950x under 80-90C
As the computer is not fully stable with the settings, it would not crash out on normal usage and the bus errors were so rare that i would get them once per month. and the only way of actually crashing the pc is to render 3-6h of my youtubevideos as a memorytest does not bring out the errors nowadays.

View attachment 196951
So this is the ram i have on my pc and i was thinking i would stuff this to the ram as the ram is not that good to begin with just to get some performance and get the pc not bluescreen when overclocked.
View attachment 196952

Albeit if you have better settings to try on the machine as its the first time i have been doing memory overclocking myself so i have no idea where this memory would go to or where to take it since ryzen 5000 settings are not on the calculator.

Based on what I see for your dram calc screen shot you are not importing your XMP into it.

You should be using Thaiphoon burner to dump the config from ram then importing that into the dram calc then choose Safe or Fast.

Here is mine this section on your screenshot its blank.