Author Topic: Ryzen 9 3950X dead cores on cold boot  (Read 3315 times)

2020-02-26, 17:12:21

Marijan

  • Active Users
  • **
  • Posts: 62
    • View Profile
Hi, as you can see in subject new Ryzen 9 3950x configuration runs with some cores inactive and speeds around 0.5 Ghz.
This happens randomly after cold boot. Sometimes it works as expected and sometimes with this problem present.
Windows 10 and all drivers, chipset and bios are updated. Memory is set on xmp profile in bios, all other parameters are default.

Configuration runs on expected temperatures and works with no problems once it boots as it should.
I haven't found much info on this problem except for this link, symptoms are the same: https://linustechtips.com/main/topic/1146384-3950x-dead-cores-on-cold-boot/

If somebody has any ideas please help :)

Full configuration:
RYZEN 9 3950X, Thermaltake water 3.0, Asus Strix x570-e gaming, 2xG-Skill TridentZ Neo 2x8GB 3600(32 gb total), Asus Strix RTX 2080Ti, Riotoro Enigma g2 psu 850w, win 10 pro

Thanks!




2020-02-26, 18:14:56
Reply #1

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12816
  • Marcin
    • View Profile
Quote
And some issues:
- I've read about it somewhere else...so did already happen to someone just can't remember where. Upon random restart (by myself), the CPU only ran at <1GHz (more like 0.5! GHz) frequency, and would only run 16 threads? After restart, it once again corrected itself.
EDIT: Solved, this is memory related issue. Changed last timing to 38 and upped voltage to 1.36. After 20 restart and cold-boots, it never materialized again. I really want to try with better memory..

Source: https://forum.corona-renderer.com/index.php?topic=21416.msg164632#msg164632
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2020-02-26, 20:55:26
Reply #2

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4769
    • View Profile
    • studio website
So, as Maru linked, this happened to me on TRX40 platform, and 3990X. So what the guy in that Linus thread considers ("taiwanese advice") the CPU being wrong.. perhaps. I'll check the thread later to see if new one solved it.

But here is my theory why that is unrelated.

- For ZEN, the IO die shares capacity together with all dies. It hosts memory controller which on Intel is usually on chipset outside of CPU. This is the reason why Ryzen & Threadrippers are so much more sensitive to memory settings related to stability.
- The XMP profile is general, and with few specific exceptions, wasn't designed or validated for ZEN platform. What can be perfectly stable on Intel CPU, isn't on Zen chips.

There are multiple ways this can manifest:
1) Doesn't even boot.
2) Crashes randomly.
3) Simply throws memory errors in MemTest (causes scientific computations to be incorrect, hence why they use ECC)
4) Our new issue...  cripples the chiplets communication.

Possible solutions:
- Set your settings manually. There is pretty complex process involving the Dram calculator and manually inputting all these settings in bios yourself. In your specific case,the 3600 MHz might be completely fine, but maybe the timings are too aggressive, voltages bit low for the platform, etc..

But this is probably not a single factor. For me this was 100perc. solution, but the issue can be two-fold. The memory settings themselves...and the current bios settings that are related to memory. For example by default, Load Balance Control is on, this is FCC regulation settings that makes BLCK=100 into 99.7, 100.3,etc... constantly changing to avoid radio signal interference. This can cause instability in overclocking...and XMP profile is overclocking, because the native JEDEC frequency of these memories is 2133-2666.

So it is quite possible that future bios updates (mainly from AGESA) can solve this, but at the moment, you can only solve it by tweaking the memory settings.

To make sure this issue for your is also memory related, disable XMP and do 20-30 cold boots. Does the issue go away ? Than it is the same.

EDIT: My post number is exactly 3950 :- D
« Last Edit: 2020-02-26, 20:59:01 by Juraj Talcik »
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2020-02-27, 12:03:33
Reply #3

Marijan

  • Active Users
  • **
  • Posts: 62
    • View Profile
Maru thanks for the link, I somehow overlooked it while searching forum...
Juraj thanks for the detailed answer, really appreciate your advise and help on forum.

This configuration is my coworkers home computer. He said that same thing is happening with memory set on auto in bios.
In the meantime he sent the computer back to the store for servicing, we'll let them know about all this Juraj posted.
Let's hope they'll solve it.

2020-02-27, 12:55:39
Reply #4

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4769
    • View Profile
    • studio website
I also have Asus Strix x570-e gaming for our 3950X build..I have no idea what he means by memory on "auto". That's not disabled XMP.
But please keep us updated, I am interested what others will find.

For what it's worth, I run 4x32GB 3200 CL16 Corsair kit stability on the 3950X, with the XMP profile. Perfectly stable. I just have shit (and mixed) memory for my 3990X for now : /.
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2020-02-27, 19:10:28
Reply #5

Marijan

  • Active Users
  • **
  • Posts: 62
    • View Profile
Sorry, not memory on "auto", I've written it wrong.
He sets XMP disabled and than memory is set to default speed which I think he said is 2133.

I have 4x16GB 2933 gskill on 2990wx with XMP profile, also works completely stable.
Would love to get my hands on 3990X, looks like real beast :).

Anyway I'll update this once we get more info.