AMD ROCm PyTorch Diagnostic Report | Generated by AI
Here’s the full diagnostic report:
✅ PyTorch + ROCm Status
| Component | Status |
|---|---|
| PyTorch (system venv) | ✅ 2.9.1+rocm installed at /opt/venv |
| torchaudio | ✅ 2.9.0 |
| torchvision | ✅ 0.24.1 |
| vLLM | ✅ 0.16.1.dev0 (ROCm 7.2.1 build) |
| APEX | ✅ 1.9.0+rocm7.2.1 |
| amdsmi | ✅ 26.2.2 |
| amd-aiter | ✅ 0.1.7 |
🖥️ GPU Hardware — 8× AMD Radeon (gfx1100 / RDNA3)
Device ID: 1002:744b → gfx1100 (AMD Radeon RX 7000 series)
VRAM each: 48 GiB (51,522,830,336 bytes)
Total VRAM: ~384 GiB
Full GPU list from lspci:
03:00.0, 23:00.0, 43:00.0, 63:00.0
83:00.0, a3:00.0, c3:00.0, e3:00.0
Plus one ASPEED BMC for display (1a03:2000).
🧠 ROCm Stack
| Component | Version |
|---|---|
| ROCm | 7.2.1 (/opt/rocm → /opt/rocm-7.2.1) |
| hipconfig | 7.2.53211 |
| Kernel driver | ROCk module 6.16.13 loaded |
| HSA Runtime | 1.18 |
⚙️ System Specs
- CPU: AMD EPYC 9334 (2× 32-core = 128 threads)
- RAM: 503 GiB
- Swap: none
📍 How to use PyTorch with ROCm
The PyTorch is in the system venv — activate it:
source /opt/venv/bin/activate
python -c "import torch; print(torch.cuda.is_available(), torch.cuda.device_count())"
# → True, 8
If your current shell can’t import torch, you’re likely running outside the venv. Just source /opt/venv/bin/activate first.
Generated by AI. Curating and sharing still takes effort.
If you find it useful, feel free to
donate.
WeChat: @lzwjava
·
X: @lzwjava ·
Say hi 👋