Stability issue after installing faster RAM (3200 MT/s → 3600 MT/s)

QuarkyPhysicist · December 29, 2022, 6:49pm

My setup

I’m running Linux: 6.0.15-300.fc37.x86_64

I have a Ryzen APU, until recently I had a pair of 3200 MT/s Corsairs installed, and it was running great.

Machine:
  Type: Desktop System: ASUS product: N/A v: N/A serial: N/A
  Mobo: ASUSTeK model: TUF GAMING B550M-PLUS (WI-FI) v: Rev X.0x
    serial: 201177404112392 UEFI: American Megatrends v: 2806 date: 10/27/2022
CPU:
  Info: 6-core model: AMD Ryzen 5 PRO 4650G with Radeon Graphics bits: 64
    type: MT MCP cache: L2: 3 MiB
  Speed (MHz): avg: 2331 min/max: 400/4308 cores: 1: 2100 2: 2207 3: 3678
    4: 2374 5: 2414 6: 2251 7: 2107 8: 2068 9: 3676 10: 2375 11: 400 12: 2325

But a couple of weeks ago, I got a pair of 3600 MT/s RAM (and double the capacity).

Memory:
  RAM: total: 23.27 GiB used: 8.88 GiB (38.2%)
  Array-1: capacity: 128 GiB slots: 4 EC: None
  Device-1: DIMM_A1 type: no module installed
  Device-2: DIMM_A2 type: DDR4 size: 16 GiB speed: 3600 MT/s
  Device-3: DIMM_B1 type: no module installed
  Device-4: DIMM_B2 type: DDR4 size: 16 GiB speed: 3600 MT/s

However I have had some lockups or crashes after my session gets longer than a few days. One very obvious change is, I consistently get stuttering and tears in YT videos (but not on Netflix). The artifacts are limited to the video area, but when this happens the rest of the desktop doesn’t show any issues.

This is how it’s setup in the BIOS:

Example lockups

While playing “STAR WARS JEDI: Fallen Order” earlier today:

Summary

Dec 29 18:09:03 kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, signaled seq=13714662, emitted seq=13714664
Dec 29 18:09:03 kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process starwarsjedifal pid 322119 thread dxvk-submit pid 322413
Dec 29 18:09:03 kernel: amdgpu 0000:07:00.0: amdgpu: GPU reset begin!
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: MODE2 reset
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: GPU reset succeeded, trying to resume
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: RAS: optional ras ta ucode is not available
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: RAP: optional rap ta ucode is not available
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: SMU is resuming...
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: SMU is resumed successfully!
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring gfx uses VM inv eng 0 on hub 0
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.0.0 uses VM inv eng 1 on hub 0
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.1.0 uses VM inv eng 4 on hub 0
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.2.0 uses VM inv eng 5 on hub 0
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.3.0 uses VM inv eng 6 on hub 0
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.0.1 uses VM inv eng 7 on hub 0
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.1.1 uses VM inv eng 8 on hub 0
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.2.1 uses VM inv eng 9 on hub 0
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.3.1 uses VM inv eng 10 on hub 0
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring kiq_2.1.0 uses VM inv eng 11 on hub 0
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring sdma0 uses VM inv eng 0 on hub 1
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring vcn_dec uses VM inv eng 1 on hub 1
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring vcn_enc0 uses VM inv eng 4 on hub 1
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring vcn_enc1 uses VM inv eng 5 on hub 1
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: ring jpeg_dec uses VM inv eng 6 on hub 1
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: recover vram bo from shadow start
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: recover vram bo from shadow done
Dec 29 18:09:04 kernel: amdgpu 0000:07:00.0: amdgpu: GPU reset(2) succeeded!
Dec 29 18:09:04 kernel: [drm:amdgpu_cs_ioctl [amdgpu]] *ERROR* Failed to initialize parser -125!

While playing Stellaris last night:

Summary

Dec 29 00:39:14 kernel: [drm:gfx_v9_0_priv_reg_irq [amdgpu]] *ERROR* Illegal register access in command stream
Dec 29 00:39:14 kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, signaled seq=61625061, emitted seq=61625063
Dec 29 00:39:14 kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process stellaris pid 1256601 thread stellaris:cs0 pid 1256602
Dec 29 00:39:14 kernel: amdgpu 0000:07:00.0: amdgpu: GPU reset begin!
Dec 29 00:39:14 kernel: amdgpu 0000:07:00.0: amdgpu: MODE2 reset
Dec 29 00:39:14 kernel: amdgpu 0000:07:00.0: amdgpu: GPU reset succeeded, trying to resume
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: RAS: optional ras ta ucode is not available
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: RAP: optional rap ta ucode is not available
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: SMU is resuming...
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: SMU is resumed successfully!
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring gfx uses VM inv eng 0 on hub 0
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.0.0 uses VM inv eng 1 on hub 0
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.1.0 uses VM inv eng 4 on hub 0
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.2.0 uses VM inv eng 5 on hub 0
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.3.0 uses VM inv eng 6 on hub 0
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.0.1 uses VM inv eng 7 on hub 0
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.1.1 uses VM inv eng 8 on hub 0
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.2.1 uses VM inv eng 9 on hub 0
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring comp_1.3.1 uses VM inv eng 10 on hub 0
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring kiq_2.1.0 uses VM inv eng 11 on hub 0
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring sdma0 uses VM inv eng 0 on hub 1
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring vcn_dec uses VM inv eng 1 on hub 1
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring vcn_enc0 uses VM inv eng 4 on hub 1
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring vcn_enc1 uses VM inv eng 5 on hub 1
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: ring jpeg_dec uses VM inv eng 6 on hub 1
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: recover vram bo from shadow start
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: recover vram bo from shadow done
Dec 29 00:39:15 kernel: amdgpu 0000:07:00.0: amdgpu: GPU reset(2) succeeded!

Example crashes

Few crashes in the last 10 days:

$ coredumpctl list -S 2022-12-20
TIME                            PID  UID  GID SIG     COREFILE EXE                                                                                 SIZE
Tue 2022-12-20 18:07:22 CET  535628 1000 1000 SIGABRT missing  /app/bin/io.github.NhekoReborn.Nheko                                                 n/a
Tue 2022-12-20 23:06:47 CET   14344 1000 1000 SIGSEGV missing  /home/user/.var/app/com.valvesoftware.Steam/.local/share/Steam/ubuntu12_32/steam   n/a
Thu 2022-12-22 18:40:43 CET 1388860 1000 1000 SIGSEGV missing  /usr/bin/ffmpeg                                                                      n/a
Thu 2022-12-22 21:07:46 CET    3212 1000 1000 SIGSEGV missing  /home/user/.var/app/com.valvesoftware.Steam/.local/share/Steam/ubuntu12_32/steam   n/a
Fri 2022-12-23 22:23:02 CET  431075 1000 1000 SIGABRT missing  /home/user/.var/app/com.mojang.Minecraft/.minecraft/launcher/minecraft-launcher    n/a
Tue 2022-12-27 00:18:16 CET  227931 1000 1000 SIGBUS  present  /usr/lib64/firefox/firefox                                                         72.2M
Tue 2022-12-27 02:44:23 CET   19930 1000 1000 SIGSEGV present  /usr/lib64/firefox/firefox                                                         26.8M
Tue 2022-12-27 19:12:55 CET    1970 1000 1000 SIGABRT present  /usr/bin/xfwm4                                                                      3.9M
Tue 2022-12-27 21:04:03 CET   19697 1000 1000 SIGSEGV present  /usr/lib64/firefox/firefox                                                         93.8M
Wed 2022-12-28 14:28:38 CET 1092307 1000 1000 SIGSEGV present  /tmp/hwi-2.1.117276974205087653776.tmp                                             74.1M
Wed 2022-12-28 14:28:38 CET 1092306 1000 1000 SIGSEGV present  /tmp/hwi-2.1.117276974205087653776.tmp                                              1.0M
Thu 2022-12-29 03:40:28 CET    2710 1000 1000 SIGABRT present  /usr/bin/emacs-28.1                                                                26.6M
Thu 2022-12-29 04:30:33 CET   79487    0    0 SIGSEGV present  /usr/bin/python3.11                                                                42.0M

Anyone has any thoughts how to fix this?

AbsolutelyFree · December 29, 2022, 8:33pm

I would definitely start with running a full round of Memtest to see if you are getting errors there. If so, try stepping the speed down slightly (3600 → 3533) and run Memtest again to see if errors persist.

alkafrazin · December 29, 2022, 10:10pm

IMO, test it at 3200 to see if the ram is stable at all.

BigBodZod · December 30, 2022, 12:35am

I would do that same AND check with the mobo vendor for any BIOS/Firmware updates.

Many times these will add better support for faster and different manufacturers memory modules.

As an example, my new Asus X670E motherboard needed to have a firmware update to properly support the faster EXPO ram kit from Corsair.

Once I did this and updated Ryzen Master then I could apply and test the EXPO timings without Windows crashing or locking up.

QuarkyPhysicist · December 30, 2022, 2:51am

@alkafrazin this particular RAM has no 3200 MT/s profile (Fury Renegade)

@BigBodZod I’m running a BIOS version from 10/27/2022 (it’s the latest on ASUS’s website). I actually had to update because I fucked up the setting when I orginally installed the new RAM, instead of picking the DOCP profile, mistakenly I had set the frequency directly, and in the process corrupting my BIOS.

As for memtest, I’m working on it. Apparently the standard way to run memtest during boot on Linux (from GRUB), doesn’t support EFI systems!

alkafrazin · December 30, 2022, 2:57am

Try Jedec, and just up the clockspeed to 3200 with timings on auto, with the voltage on 1.3 or 1.35.

AbsolutelyFree · December 30, 2022, 3:35am

You can download the free version of Memtest86, install it to a USB drive, and boot from the USB drive.

anon7678104 · December 30, 2022, 5:37am

find the make and model of the ic’s used for your ram.
download ryzen dram calculator and enter your ram spec and ic type.
hit calculate safe.

take a photo of the screen
go to the next tab and do the same, enter the specs and take a photo of the results.

go into bios/uefi
now manually set your ram volts to the xmp rating (typically 1.35 but check to be sure).
then enter the mt/s rate and all the timings you can match to the ones in the photos.
not everything may be listed in your boards uefi so just enter as much of the info as it will allow.

reboot. let it train and hope for the best…

jode · December 30, 2022, 6:30am

I think the Ryzen APUs have issues with RAM > DDR4-3200.

I was having similar issues installing a Ryzen 5700G into a X570 mobo that I have run successfully with DDR4-4000 @CL18 and a Ryzen 5900X.

Issues went away when slowing down to DDR-3200.

anon7678104 · December 30, 2022, 6:51am

^^ good catch…

in that case do as before but set the ram MT/s to 3200 you should get some real low latencies in return for the downgrade in MT/s… which can be a worth while trade off in some games.

or if you have a dedicated gpu also turn the graphics off on the chip. it might allow you to run the faster ram that way.

QuarkyPhysicist · December 30, 2022, 3:38pm

I managed to get both the FOSS (memtest86+) and the proprietary version (memtest86) working with a Ventoy USB drive (great project btw, and it’s a one-person show!).

I have errors in the first pass already. I checked the AMD product page, it does indeed mention “upto 3200” , @jode is right. I should have checked before purchasing.

I tried picking the lower clocked profile (3000 @ CL15, my old RAM was 3200 @ CL16). Tests are running, so far no errors after completing one pass. So for now this works, but the throughput is same as my old RAM:

16 GB/s running at 3000 @ CL15
18 GB/s running at 3600 @ CL16 (with errors)
16 GB/s running at 3200 @ CL16 (old RAM)

@anon7678104 I’m not sure I can follow your instructions. The RAM is this one. Short of taking the heatsinks off, I don’t know how to find the exact IC. Also, I don’t have any Windows machine around to run the Ryzen DRAM calculator (is it this one?). Would it work in a VM? I could do that.

anon7678104 · December 30, 2022, 4:04pm

on the dropdown theres a list of ic types by manufacturer… look up which you have (it may take some digging as it didnt show up straight way when i googled the part number).

QuarkyPhysicist · December 30, 2022, 4:46pm

I searched around, the only reference to the specific die I found was in this article: SK Hynix D-die.

But I decided to return these sticks. This machine also serves as my home workstation, so can’t really spend the time rebooting and tuning. I’ll just get sticks that match my current pair. I need the capacity bump more than the speed.

Mark1 · December 30, 2022, 10:30pm

Memtest could also be installed and run from within his Linux installation. On Ubuntu that works well.

QuarkyPhysicist · December 30, 2022, 10:39pm

For older versions of memtest86+ on EFI systems it’s not possible; see this.

I was running with 3200 kit for over a year, upgrading to 3600 caused the instability. As mentioned earlier in the thread, APUs only support upto 3200, unfortunately I didn’t know this before purchasing my kit. No harm done, I have initiated a return, and bought a matching set of sticks to supplement the capacity.

Mark1 · December 30, 2022, 10:45pm

I was thinking of this tool: Ubuntu – Details of package memtester in jammy - I wrote the wrong name there.

QuarkyPhysicist · December 30, 2022, 10:50pm

I tried that as well, didn’t really give me anything useful. But it’s possible I used it wrong. I was providing the size I should test, but I kept getting an error similar to “too large, trying a smaller size” until I lowered it to 1G. That seemed a bit ridiculous since my kit is 2x16G. So I didn’t bother investigating further.

Mark1 · December 30, 2022, 11:10pm

You probably need to run it using sudo, or as root.

I usually ran the test for 75% of the installed total memory size.

QuarkyPhysicist · December 30, 2022, 11:13pm

Ah nice! I just tried it, running as root gets it to work! Thanks

Zaf9670 · January 5, 2023, 4:31pm

I see you might have the solution but I would also use the QVL in the future for supported RAM. I’m not sure what happens if you enable beyond 3200 and the APU does not support it.

With the AMD chiplets and all these OC memory speeds the QVL is more important IMHO but others may disagree. I think the link for your motherboard is below.