ROCm: Blender Cycles engine causes "Memory access fault by GPU node-1"

Problem Description

Error: Blender crashes with Memory access fault by GPU node-1 (Agent handle: 0x7c1aea898400) on address 0x5fd000. Reason: Page not present or supervisor privilege.

Operating System

6.7.4-2-MANJARO Linux

CPU

AMD Ryzen 9 5950X

GPU

AMD Radeon Pro W7900

ROCm Version

ROCm 6.0.0

ROCm Component

No response

Steps to Reproduce

Reproducing error: Just opening Blender (tested 4.0-4.1), switching Render Engine to Cycles (Feature Set: Supported, Device: GPU compute) Activate Viewport shading (and maybe rotate the camera around the default cube a few times) causes the crash. Also trying to render anything instantly crashes Blender.

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

ROCk module is loaded

HSA System Attributes

Runtime Version: 1.1 System Timestamp Freq.: 1000.000000MHz Sig. Max Wait Duration: 18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count) Machine Model: LARGE
System Endianness: LITTLE
Mwaitx: DISABLED DMAbuf Support: YES

==========
HSA Agents


Agent 1


Name: AMD Ryzen 9 5950X 16-Core Processor Uuid: CPU-XX
Marketing Name: AMD Ryzen 9 5950X 16-Core Processor Vendor Name: CPU
Feature: None specified
Profile: FULL_PROFILE
Float Round Mode: NEAR
Max Queue Number: 0(0x0)
Queue Min Size: 0(0x0)
Queue Max Size: 0(0x0)
Queue Type: MULTI
Node: 0
Device Type: CPU
Cache Info:
L1: 32768(0x8000) KB
Chip ID: 0(0x0)
ASIC Revision: 0(0x0)
Cacheline Size: 64(0x40)
Max Clock Freq. (MHz): 3400
BDFID: 0
Internal Node ID: 0
Compute Unit: 32
SIMDs per CU: 0
Shader Engines: 0
Shader Arrs. per Eng.: 0
WatchPts on Addr. Ranges:1
Features: None Pool Info:
Pool 1
Segment: GLOBAL; FLAGS: FINE GRAINED
Size: 131808628(0x7db3d74) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: TRUE
Pool 2
Segment: GLOBAL; FLAGS: KERNARG, FINE GRAINED Size: 131808628(0x7db3d74) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: TRUE
Pool 3
Segment: GLOBAL; FLAGS: COARSE GRAINED
Size: 131808628(0x7db3d74) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: TRUE
ISA Info:


Agent 2


Name: gfx1100
Uuid: GPU-9e8b159eef31d025
Marketing Name: AMD Radeon Pro W7900
Vendor Name: AMD
Feature: KERNEL_DISPATCH
Profile: BASE_PROFILE
Float Round Mode: NEAR
Max Queue Number: 128(0x80)
Queue Min Size: 64(0x40)
Queue Max Size: 131072(0x20000)
Queue Type: MULTI
Node: 1
Device Type: GPU
Cache Info:
L1: 32(0x20) KB
L2: 6144(0x1800) KB
L3: 98304(0x18000) KB
Chip ID: 29768(0x7448)
ASIC Revision: 0(0x0)
Cacheline Size: 64(0x40)
Max Clock Freq. (MHz): 1760
BDFID: 3584
Internal Node ID: 1
Compute Unit: 96
SIMDs per CU: 2
Shader Engines: 6
Shader Arrs. per Eng.: 2
WatchPts on Addr. Ranges:4
Coherent Host Access: FALSE
Features: KERNEL_DISPATCH Fast F16 Operation: TRUE
Wavefront Size: 32(0x20)
Workgroup Max Size: 1024(0x400)
Workgroup Max Size per Dimension: x 1024(0x400)
y 1024(0x400)
z 1024(0x400)
Max Waves Per CU: 32(0x20)
Max Work-item Per CU: 1024(0x400)
Grid Max Size: 4294967295(0xffffffff)
Grid Max Size per Dimension: x 4294967295(0xffffffff)
y 4294967295(0xffffffff)
z 4294967295(0xffffffff)
Max fbarriers/Workgrp: 32
Packet Processor uCode:: 528
SDMA engine uCode:: 19
IOMMU Support:: None
Pool Info:
Pool 1
Segment: GLOBAL; FLAGS: COARSE GRAINED
Size: 47169536(0x2cfc000) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: FALSE
Pool 2
Segment: GLOBAL; FLAGS: EXTENDED FINE GRAINED Size: 47169536(0x2cfc000) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: FALSE
Pool 3
Segment: GROUP
Size: 64(0x40) KB
Allocatable: FALSE
Alloc Granule: 0KB
Alloc Alignment: 0KB
Accessible by all: FALSE
ISA Info:
ISA 1
Name: amdgcn-amd-amdhsa–gfx1100
Machine Models: HSA_MACHINE_MODEL_LARGE
Profiles: HSA_PROFILE_BASE
Default Rounding Mode: NEAR
Default Rounding Mode: NEAR
Fast f16: TRUE
Workgroup Max Size: 1024(0x400)
Workgroup Max Size per Dimension: x 1024(0x400)
y 1024(0x400)
z 1024(0x400)
Grid Max Size: 4294967295(0xffffffff)
Grid Max Size per Dimension: x 4294967295(0xffffffff)
y 4294967295(0xffffffff)
z 4294967295(0xffffffff)
FBarrier Max Size: 32
*** Done ***

Additional Information

OS: NAME=“Manjaro Linux”

CPU: model name : AMD Ryzen 9 5950X 16-Core Processor

GPU: Name: AMD Ryzen 9 5950X 16-Core Processor Marketing Name: AMD Ryzen 9 5950X 16-Core Processor Name: gfx1100
Marketing Name: AMD Radeon Pro W7900
Name: amdgcn-amd-amdhsa–gfx1100

Mobo: ASUSTeK model: ROG STRIX X570-E GAMING WIFI II

Graphics: Device-1: AMD Navi 31 [Radeon Pro W7900] driver: amdgpu v: kernel Display: x11 server: X.Org v: 21.1.11 with: Xwayland v: 23.2.4 driver: X: loaded: amdgpu unloaded: modesetting,radeon dri: radeonsi gpu: amdgpu resolution: 3840x2160~144Hz API: EGL v: 1.5 drivers: radeonsi,swrast platforms: x11,surfaceless,device API: OpenGL v: 4.6 compat-v: 4.5 vendor: amd mesa v: 23.3.5-manjaro1.1 renderer: AMD Radeon Pro W7900 (radeonsi navi31 LLVM 16.0.6 DRM 3.57 6.7.4-2-MANJARO) API: Vulkan v: 1.3.276 drivers: amd surfaces: xcb,xlib

RAM (DDR4) Memory: total: 128 GiB

PSU: 1000W

Case: EATX with 9 Fans

About this issue

  • Original URL
  • State: closed
  • Created 3 years ago
  • Comments: 37 (2 by maintainers)

Most upvoted comments

@NerosTie Internal ticket has been created to investigate this issue. Thanks!

You can still reproduce the crash with a slight caveat. The reproducing step with Docker is as follows:

  1. Pull the rocm/rocm-terminal:6.1 docker image.
  2. Enter the docker image: podman run -it --device=/dev/kfd --device=/dev/dri --group-add=video --ipc=host --cap-add=SYS_PTRACE --security-opt seccomp=unconfined rocm/rocm-terminal:6.1. I personally use podman, if you use docker replace podman with sudo docker.
  3. Create a work directory for convenience. mkdir work && cd work
  4. Install runtime dependencies for Blender: sudo apt-get update && sudo apt-get install build-essential git git-lfs subversion cmake libx11-dev libxxf86vm-dev libxcursor-dev libxi-dev libxrandr-dev libxinerama-dev libegl-dev libwayland-dev wayland-protocols libxkbcommon-dev libdbus-1-dev linux-libc-dev unzip libsm6 libxext6 -y. (yes, not all of these dependencies are needed, I’m just being safe here)
  5. Fetch the Blender 4.1 binary: curl -O https://download.blender.org/release/Blender4.1/blender-4.1.0-linux-x64.tar.xz
  6. Extract Blender: tar xf blender-4.1.0-linux-x64.tar.xz. You should now have a blender-4.1.0-linux-x64 directory in you current directory.
  7. Fetch the BMW27 blender project for testing: curl -O https://download.blender.org/demo/test/BMW27.blend.zip
  8. unzip BMW27.blend.zip. You should now have a BMW27.blend file in your current directory.
  9. Render the file. blender-4.1.0-linux-x64/blender -b BMW27.blend -f 0 -- --cycles-device HIP. Note that by default Blender will use your GPU with id 0 to render; if you want to use another GPU, run export HIP_VISIBLE_DEVICES=<gpu-id-you-want-to-use> before rendering with Blender. You can find the available GPU ids by running rocm-smi (under the Device column).
  10. You should now almost immediately get Memory access fault by GPU node-1 (Agent handle: 0x7f5f2af98000) on address 0x7f5afa7cb000. Reason: Page not present or supervisor privilege.

The caveat is that this issue is probably on the compiler side, not ROCm’s runtime. Blender ships pre-compiled GPU fatbins that is compiled with some ROCm version that I don’t know (I suspect the fatbins that come with 4.1 are compiled with ROCm 6.0, but even the Blender folks aren’t sure). To verify that it’s probably a compiler issue, continue:

  1. Clone Blender. To save space and ensure consistency, we’re just going to clone the 4.1.0 version: git clone https://projects.blender.org/blender/blender.git --depth 1 --branch v4.1.0
  2. Enter the Blender folder and compile the fatbins with ROCm 6.1: pushd blender && hipcc --offload-arch=$arch --genco intern/cycles/kernel/device/hip/kernel.cpp -D CCL_NAMESPACE_BEGIN= -D CCL_NAMESPACE_END= -D HIPCC -I intern/cycles/kernel/.. -I intern/cycles/kernel/device/hip -ffast-math -o kernel_$arch.fatbin && popd. Replace $arch with the architecture of the GPU you want to run on, e.g. gfx900. If you want to compile against multiple architectures, you must repeat step 12 and 13 for each architecture.
  3. Put the blender/kernel_$arch.fatbin file at blender-4.1.0-linux-x64/4.1/scripts/addons/cycles/lib/kernel_$arch.fatbin. Once again, replace $arch with your desired GPU architecture.
  4. Now run step 9 again. Blender shouldn’t crash anymore.

If you’re unable to reproduce the crash with the BMW27 example, try any one of the demo files under the “Cycles” section on the Blender demos page. One common test file we used was the Classroom example.

Running HSA_ENABLE_SDMA=0 blender-4.2 gives me the same result.

Memory access fault by GPU node-1 (Agent handle: 0x7c32f76c4e00) on address 0x7c315dde2000. Reason: Page not present or supervisor privilege.
Aborted (core dumped)