ROCm Configuration for AMD Radeon RX 7700 XT

December 6, 2025 ยท View on GitHub

Generated: 2025-12-06

ROCm Installation

PropertyValue
ROCm Version6.3.0-39
ROCM_PATH/opt/rocm
Installed Packages25 packages

Installed ROCm Packages

PackageVersionDescription
rocm-core6.3.0.60300-39ROCm Runtime software stack
rocm-llvm18.0.0.24455.60300-39ROCm core compiler
rocm-device-libs1.0.0.60300-39Device libraries
hsa-rocr1.14.0.60300-39HSA Runtime for ROCm
rocblas4.3.0.60300-39BLAS library for ROCm
rocsolver3.27.0.60300-39Numerical linear algebra
rocalution3.2.1.60300-39Sparse linear systems
rccl2.21.5.60300-39Communication Collectives
rocm-smi-lib7.4.0.60300-39System Management Interface
rocminfo6.4.3-1System Info Reporting
comgr2.8.0.60300-39Code Object Manager
rocm-opencl-icd5.7.1-6OpenCL ICD runtime

Critical Environment Variables

For proper GPU detection with gfx1101, set these environment variables:

# Required for gfx1101 (Navi 32) GPU compatibility
export HSA_OVERRIDE_GFX_VERSION=11.0.1

# ROCm path
export ROCM_PATH=/opt/rocm

# GPU visibility (optional, for multi-GPU systems)
export HIP_VISIBLE_DEVICES=0
export ROCR_VISIBLE_DEVICES=0

PyTorch ROCm Configuration

PropertyValue
PyTorch Version2.9.1+rocm6.3
Installation Location~/.local/lib/python3.13/site-packages
ROCm/HIP AvailableYes
Device Count1
Device NameAMD Radeon Graphics
Available VRAM12.0 GB

PyTorch Dependencies

  • pytorch-triton-rocm 3.5.1
  • torchaudio
  • torchvision
  • accelerate
  • optimum

PyTorch GPU Test

import torch

print(f"PyTorch: {torch.__version__}")
print(f"ROCm/HIP available: {torch.cuda.is_available()}")
print(f"Device count: {torch.cuda.device_count()}")
print(f"Device name: {torch.cuda.get_device_name(0)}")
print(f"VRAM: {torch.cuda.get_device_properties(0).total_memory / 1024**3:.1f} GB")

Known Compatibility Notes

gfx1101 Considerations

The AMD Radeon RX 7700 XT (gfx1101) is part of the RDNA 3 architecture. Key notes:

  1. HSA Override: The HSA_OVERRIDE_GFX_VERSION=11.0.1 environment variable is often required for proper GPU detection in ROCm applications.

  2. xformers: Not compatible with ROCm. Disable with:

    export XFORMERS_DISABLED=1
export DISABLE_XFORMERS=1

  3. CUDA Compatibility Layer: ROCm provides HIP which translates CUDA calls. In PyTorch, torch.cuda functions work with ROCm GPUs.

  4. Docker/Container Support: For containerized workloads, use --device=/dev/kfd --device=/dev/dri and pass ROCm-related environment variables.

Kernel Module Status

ROCk module is loaded
amdgpu kernel driver in use

HSA Agent Detection

The GPU is properly detected as HSA Agent 2:

  • Name: gfx1101
  • Type: GPU
  • Feature: KERNEL_DISPATCH
  • Profile: BASE_PROFILE