[Issue]: Dual Radeon AI PRO R9700 – intermittent full-system resets during ROCm inference, Data Fabric Sync Flood 0x08000800
Nadie ha tomado este issue todavía.
Evaluación
- Dificultad
- 4/5
- Tiempo estimado
- 3-5 días
- Aptitud para principiantes
- 25/100
- Tipo de issue
- Error
- Claridad
- Necesita aclaración
- Estado de actividad
- Activo
- Stack tecnológico
- linux
Línea de trabajo
Start by reviewing the attached kernel logs, PCIe dumps, and ROCm diagnostics, then compare the reported reset messages with the dual-GPU configuration and workload details. The issue names no source files or tests; progress would mean identifying whether the failures point to a driver, GPU, or PCIe problem and documenting useful diagnostics or a confirmed workaround.
Escrito por el modelo de indexación a partir del texto del issue.
Descripción
Problem Description
Dear AMD ROCm Engineering Team,
I am experiencing intermittent hard system resets during local LLM inference on a workstation equipped with two Radeon AI PRO R9700 GPUs.
These are complete machine resets, not just application crashes. The system may restart without a preceding kernel panic or a recoverable GPU reset.
System configuration
CPU: AMD Ryzen 9 7900
Motherboard: ASUS ProArt X870E-CREATOR WIFI
BIOS: 2503, AGESA 1.3.0.1d
GPU 1: Radeon AI PRO R9700 32 GB, 0000:03:00.0
GPU 2: Radeon AI PRO R9700 32 GB, 0000:07:00.0
GPU architecture: gfx1201
Operating system: Ubuntu Linux
ROCm driver: 7.1.9.31600000
amdgpu-dkms: 7.1.9.31600000
amdgpu packages: 26.14.26140000
Workload: llama.cpp / llama-server with local Qwen LLMs, including Qwen3-8B and a 27B model
Observed failure
The system resets unexpectedly during GPU-accelerated inference. The problem has occurred during model processing and other GPU-intensive operations, but is not strictly limited to the largest prompts.
After reboot, the kernel reports:
x86/amd: Previous system reset reason [0x08000800]:
an uncorrected error caused a data fabric sync flood event
In earlier failure logs, I have also observed:
uncorrectable hardware error (ERREVENT_ATHUB_INTERRUPT)
device lost from bus!
Some previous logs also contained high GPU UMC error counts and subsequent GPU initialization failures. These messages were not necessarily captured immediately before every reset.
Mitigations already attempted
Disabled PCIe ASPM via pcie_aspm=off
Disabled GFXOFF for both discrete GPUs
Limited each GPU to 240 W (1200W power supply)
Selected the COMPUTE power profile
Set PCIe runtime power control to on
Bios updates
Tested with EXPO disabled
The full-system resets have continued despite these changes.
Additional PCIe observation
The first CPU PCIe root port negotiates Gen5 x8, while the second currently negotiates Gen4 x8 even though its target speed is Gen5:
00:01.1: 32 GT/s x8
00:01.3: 16 GT/s x8
The second GPU endpoint itself reports a Gen5 x16 link to its downstream PCIe switch.
I am reporting the link negotiation issue separately to ASUS. I do not know whether it is related to the resets.
Questions
Are Data Fabric Sync Flood resets or ATHUB errors known with dual Radeon AI PRO R9700 configurations?
Are there known stability issues with ROCm 7.1.3 and llama.cpp inference on gfx1201?
Could the earlier ATHUB and UMC errors indicate a GPU, PCIe, memory-management or driver issue?
What additional logs or diagnostic tests would help isolate the failure?
Is there a recommended driver version, firmware update or workaround for this configuration?
I can provide kernel logs, PCIe capability/status dumps, ROCm diagnostics and the llama-server configuration.
Thank you for investigating.
Operating System
Ubuntu 26.04.1 LTS (Resolute Raccoon)
CPU
AMD Ryzen 9 7900 12-Core Processor
GPU
2 x AMD Ryzen 9 7900 12-Core Processor
ROCm Version
7.1.3.31500000 / 7.1.9.31600000
Installation Method
Built from source
Installed ROCm Packages / Versions
:$ apt list --installed 2>/dev/null | grep -iE "rocm|hip|amdgpu"pre4-32424980240 amd64 [Installiert,lokal]
amdgpu-core/now 1:26.14.26140000-2390957.26.04 all [Installiert,aktualisierbar auf: 1:26.16.26160000-2407865.26.04]
amdgpu-dkms-firmware/now 1:31.50.0.0.31500000-2390945.26.04 all [Installiert,aktualisierbar auf: 1:31.60.0.0.31600000-2407772.26.04]
amdgpu-dkms/resolute,resolute,resolute,resolute,now 1:7.1.9.31600000-2407772.26.04 all [installiert]
amdgpu-install/resolute,resolute,resolute,resolute,now 31.60.0.0.31600000-2407772.26.04 all [installiert]
amdgpu-lib/now 1:26.14.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:26.16.26160000-2407865.26.04]
amdgpu-multimedia/now 1:26.14.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:26.16.26160000-2407865.26.04]
amdrocm-amdsmi10.0/now 10.0.0
amdrocm-base10.0/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-blas-host10.0/now 10.0.0
amdrocm-blas10.0-gfx1201/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-ck10.0/now 10.0.0
amdrocm-core10.0-gfx1201/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-decode10.0/now 10.0.0
amdrocm-dnn-host10.0/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-dnn10.0-gfx1201/now 10.0.0
amdrocm-fft-host10.0/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-fft10.0-gfx1201/now 10.0.0
amdrocm-hipfile10.0/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-hipify10.0/now 10.0.0
amdrocm-jpeg10.0/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-llvm10.0/now 10.0.0
amdrocm-math-common10.0/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-opencl10.0/now 10.0.0
amdrocm-opencl/now 10.0.0pre4-32424980240 amd64 [Installiert,aktualisierbar auf: 10.1.0-3]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-profiler-base10.0/now 10.0.0
amdrocm-rand-host10.0/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-rand10.0-gfx1201/now 10.0.0
amdrocm-rccl-host10.0/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-rccl10.0-gfx1201/now 10.0.0
amdrocm-rocshmem10.0/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-runtime10.0/now 10.0.0
amdrocm-runtime/now 10.0.0pre4-32424980240 amd64 [Installiert,aktualisierbar auf: 10.1.0-3]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-solver-host10.0/now 10.0.0
amdrocm-solver10.0-gfx1201/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-sparse-host10.0/now 10.0.0
amdrocm-sparse10.0-gfx1201/now 10.0.0pre4-32424980240 amd64 [Installiert,lokal]pre4-32424980240 amd64 [Installiert,lokal]
amdrocm-sysdeps10.0/now 10.0.0
amdrocm10.0-gfx1201/now 10.0.0~pre4-32424980240 amd64 [Installiert,lokal]
hip-utils/resolute,now 7.1.0-0ubuntu2 amd64 [Installiert,automatisch]
hipcc/resolute,now 7.1.1+dfsg-0ubuntu1 amd64 [Installiert,automatisch]
hipify-perl/resolute,now 7.1.1+dfsg-0ubuntu1 all [Installiert,automatisch]
libamdhip64-7/resolute,now 7.1.0-0ubuntu2 amd64 [Installiert,automatisch]
libamdhip64-dev/resolute,now 7.1.0-0ubuntu2 amd64 [Installiert,automatisch]
libdrm-amdgpu-amdgpu1/now 1:2.4.131.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:2.4.134.26160000-2407865.26.04]
libdrm-amdgpu-common/now 1.0.0.26140000-2390957.26.04 all [Installiert,aktualisierbar auf: 1.0.0.26160000-2407865.26.04]
libdrm-amdgpu1/resolute,now 2.4.131-1 amd64 [Installiert,automatisch]
libdrm2-amdgpu/now 1:2.4.131.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:2.4.134.26160000-2407865.26.04]
libegl1-amdgpu-mesa-drivers/now 1:26.1.3.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:26.2.0.26160000-2407865.26.04]
libegl1-amdgpu-mesa/now 1:26.1.3.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:26.2.0.26160000-2407865.26.04]
libgbm1-amdgpu/now 1:26.1.3.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:26.2.0.26160000-2407865.26.04]
libgl1-amdgpu-mesa-dri/now 1:26.1.3.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:26.2.0.26160000-2407865.26.04]
libgl1-amdgpu-mesa-glx/now 1:26.1.3.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:26.2.0.26160000-2407865.26.04]
libhipblas-common-dev/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
libhipblas-dev/resolute,now 7.1.0-0ubuntu5 amd64 [Installiert,automatisch]
libhipblas3-tests-data/resolute,now 7.1.0-0ubuntu5 all [Installiert,automatisch]
libhipblas3-tests/resolute,now 7.1.0-0ubuntu5 amd64 [Installiert,automatisch]
libhipblas3/resolute,now 7.1.0-0ubuntu5 amd64 [Installiert,automatisch]
libhipcub-dev/resolute,now 7.1.0-0ubuntu3 all [Installiert,automatisch]
libhipcub-tests/resolute,now 7.1.0-0ubuntu3 amd64 [Installiert,automatisch]
libhipfft-dev/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
libhipfft0-tests/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
libhipfft0/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
libhipfftw0/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
libhiprand-dev/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
libhiprand1-tests/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
libhiprand1/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
libhiprtc-builtins7/resolute,now 7.1.0-0ubuntu2 amd64 [Installiert,automatisch]
libhiprtc7/resolute,now 7.1.0-0ubuntu2 amd64 [Installiert,automatisch]
libhipsolver-dev/resolute,now 7.1.0-0ubuntu4 amd64 [Installiert,automatisch]
libhipsolver-fortran-dev/resolute,now 7.1.0-0ubuntu4 amd64 [Installiert,automatisch]
libhipsolver-fortran1/resolute,now 7.1.0-0ubuntu4 amd64 [Installiert,automatisch]
libhipsolver1-tests-data/resolute,now 7.1.0-0ubuntu4 all [Installiert,automatisch]
libhipsolver1-tests/resolute,now 7.1.0-0ubuntu4 amd64 [Installiert,automatisch]
libhipsolver1/resolute,now 7.1.0-0ubuntu4 amd64 [Installiert,automatisch]
libhipsparse-dev/resolute,now 7.1.0-0ubuntu1 amd64 [Installiert,automatisch]
libhipsparse4-tests/resolute,now 7.1.0-0ubuntu1 amd64 [Installiert,automatisch]
libhipsparse4/resolute,now 7.1.0-0ubuntu1 amd64 [Installiert,automatisch]
libllvm22.1-amdgpu/now 1:22.1.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:22.1.26160000-2407865.26.04]
librocm-dbgapi0/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
librocm-smi-dev/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
librocm-smi64-7/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
mesa-amdgpu-libgallium/now 1:26.1.3.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:26.2.0.26160000-2407865.26.04]
mesa-amdgpu-va-drivers/now 1:26.1.3.26140000-2390957.26.04 amd64 [Installiert,aktualisierbar auf: 1:26.2.0.26160000-2407865.26.04]
rocm-dev/resolute,now 7.1.0-0ubuntu6 amd64 [Installiert,automatisch]
rocm-device-libs-21/resolute,now 7.1.1+dfsg-0ubuntu1 amd64 [Installiert,automatisch]
rocm-tests/resolute,now 7.1.0-0ubuntu6 amd64 [Installiert,automatisch]
rocm/resolute,now 7.1.0-0ubuntu6 amd64 [installiert]
rocminfo/resolute,now 7.1.1-0ubuntu1 amd64 [Installiert,automatisch]
whiptail/resolute,now 0.52.25-1ubuntu3 amd64 [Installiert,automatisch]
ROCm Component
No response
Steps to Reproduce
AI worloads, it does not matter with olama /llama / comfy UI
(Optional for Linux users) Output of rocminfo --support
rocminfo --support output
Paste output here
Additional Information
kernel.txt
kernel-cmdline.txt
kernel-current-boot.txt
kernel-previous-boot.txt
lspci-full.txt
pcie-tree.txt
README-anonymization.txt
rocminfo.txt
rocm-smi.txt
ubuntu-version.txt
amd-packages.txt
bios.txt
dkms.txt
- Lenguaje dominante
- C
- Estrellas
- 470
- Forks
- 150
- Métricas de merge de PR
- Sin PR fusionados en 30 d
Preparar el entorno
Este proyecto no incluye contenedor de desarrollo, Dockerfile ni guía de contribución, así que la configuración corre por tu cuenta: empieza por su README y consulta nuestra guía para la primera contribución para los pasos generales.
Primeros pasos
- Lee el issue completo y luego la guía de contribución del proyecto.
- Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
- Haz un fork del repositorio y trabaja en una rama.
- Abre un pull request que haga referencia al número del issue.
Más de ROCm/amdgpu
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 74/100
-
Dificultad 5/5 Más de una semana Aptitud para principiantes 30/100
-
Dificultad 5/5 Más de una semana Aptitud para principiantes 8/100
-
amdgpu unified RAS: GFX correctable errors aggregated by MP5 firmware are dropped on MI450 (no GFX error count, no CPER)Posiblemente ocupada @tpatward-amd la tomó hace 16 días. Abierto
Dificultad 4/5 3-5 días Aptitud para principiantes 28/100
-
Dificultad 4/5 3-5 días Aptitud para principiantes 48/100
Todos los issues de ROCm/amdgpu
Issues similares
-
enhancement ui
Dificultad 1/5 Menos de una hora Aptitud para principiantes 72/100
ChupathingyCE/chupathingyce#485 ·
Los mantenedores suelen responder en 1 día
-
Need to correct all URLsAbierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 65/100
Los mantenedores suelen responder en 1 día
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 65/100
JustAResearcher/gpu-unlock-ada#98 ·
Los mantenedores suelen responder en 1 día
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 79/100
FujiNetWIFI/fujinet-firmware#2216 ·
Los mantenedores suelen responder en 1 día
-
bug
Dificultad 2/5 1-3 horas Aptitud para principiantes 84/100
Los mantenedores suelen responder en 2 días