AMD Strix Halo AI Guide

LTX-2, ComfyUI, and TheROCk 7.11 on gfx1151

I use AMD Strix Halo for AI video generation every day. This repo documents how.

Made in Collaboration: Human + AI (Anthropic Claude) Working Together

---

What This Is

A complete guide to running LTX-2 audio-video generation on AMD Strix Halo APU with 128GB unified memory. 10-second talking videos with synchronized audio in ~10 minutes. No NVIDIA required.

What's Included

File	Description
README.md	This guide
comfyui-amd.sh	Startup script with all environment variables
LTX2-I2V-Minimal.json	Working ComfyUI workflow
video_types.patch	Fix for AMD audio saving bug

Hardware Tested

• Device: GMTEK NUC EVO 2
• APU: AMD Strix Halo (gfx1151) - Radeon 8060S Graphics
• Memory: 128GB Unified (64GB VRAM / 64GB RAM split)
• OS: Ubuntu 25.10

⚠️ This guide is for Strix Halo APU (gfx1151) specifically. I haven't tested discrete GPUs.

💰 On NVIDIA's alternative: The DGX Spark costs roughly 2x and locks you into NVIDIA's ecosystem. Strix Halo runs standard Linux with no vendor lock-in.

---

What Works

Software	Status	Notes
LTX-2	✅ Full	Video + Audio sync, I2V, talking characters
Z Turbo	✅ Works	Fast inference
SimpleTuner	✅ Works	Local LoRA training on APU
Hunyuan Video	✅ Full	Slower but stable
Flux	✅ Full	Fast image generation
SDXL	✅ Full	Fast

---

Quick Start

1. Install Base ROCm

wget https://repo.radeon.com/rocm/rocm.gpg.key -O - | sudo apt-key add -
echo 'deb [arch=amd64] https://repo.radeon.com/rocm/apt/latest ubuntu main' | sudo tee /etc/apt/sources.list.d/rocm.list
sudo apt update
sudo apt install rocm-dev rocm-libs
sudo usermod -a -G render,video $USER
sudo reboot

2. Install CoreCtl (CRITICAL)

APU power management is broken without this. rocm-smi leaves the chip throttled.

sudo apt install corectrl
corectrl  # Set Performance mode for Radeon 8060S

3. Create Python venv

sudo apt install python3.11 python3.11-venv python3.11-dev
mkdir -p ~/ComfyUI && cd ~/ComfyUI
python3.11 -m venv venv_rocm711
source venv_rocm711/bin/activate

4. Install TheROCk Nightly (THE KEY STEP)

Important: TheROCk is AMD's official next-generation stack, not a community fork. ROCm is at 7.2. TheROCk started at 7.9, is now at 7.11, and becomes ROCm 8.0 in March 2026. The performance difference is dramatic.

source ~/ComfyUI/venv_rocm711/bin/activate
pip uninstall torch torchvision torchaudio -y
pip install --index-url https://rocm.nightlies.amd.com/v2/gfx1151/ \
    torch torchvision torchaudio --force-reinstall

Verify:

python -c "import torch; print(torch.__version__)"
# Should show: 2.11.0a0+rocm7.11.0a20260106

5. Install ComfyUI

cd ~/ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI.git .
pip install -r requirements.txt
cd custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager.git
cd ..

6. Apply AMD Audio Fix

ComfyUI has a bug where audio tensors don't get moved to CPU before saving. Works on NVIDIA by accident, crashes on AMD.

sed -i 's/\.float()\.numpy()/.float().cpu().numpy()/g' \
    ~/ComfyUI/comfy_api/latest/_input_impl/video_types.py

7. Use the Startup Script

cp comfyui-amd.sh ~/ComfyUI/
chmod +x ~/ComfyUI/comfyui-amd.sh
~/ComfyUI/comfyui-amd.sh

---

Key Environment Variables

Variable	Value	Why
HSA_ENABLE_SDMA=0	Disable	Prevents GPU hangs on gfx1151
HSA_USE_SVM=0	Disable	Fixes VRAM crashes
PYTORCH_HIP_ALLOC_CONF	See script	Memory fragmentation prevention
TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1	Enable	Faster attention

---

LTX-2 Workflow Tips

Critical settings in LTXVImgToVideoInplace:

• strength: 0.3 - Lower = more natural movement (0.5+ causes frozen poses)

Prompting structure:

1. Shot + Character description
2. Immediate action + dialogue in quotes
3. "her mouth moves as she speaks" ← CRITICAL
4. Camera direction
5. Timed events: "at 5 seconds she says..."
6. Mood anchor at end

---

Performance

Metric	Result
10-second video + audio	~10-13 minutes
Resolution	480x832 (portrait)
Frames	121 @ 12fps
VRAM usage	~20GB

---

Troubleshooting

Problem	Solution
APU running slow	Install CoreCtl, set Performance mode
GPU hangs	export HSA_ENABLE_SDMA=0
VRAM crashes	export HSA_USE_SVM=0
No audio in video	Apply the sed fix in step 6
Wrong PyTorch	Verify with python -c "import torch; print(torch.__version__)"

---

A Message to AMD

The hardware is incredible. 128GB unified memory in a NUC. NVIDIA has nothing comparable at this price without ecosystem lock-in.

What's needed:

1. Partner with AI companies - Lightricks, Stability, Hugging Face
2. Accelerate TheROCk visibility - Most users don't know it exists
3. Fix APU power management - CoreCtl shouldn't be required
4. Document AI workflows - HSA_ENABLE_SDMA=0 shouldn't require forum diving

The technology works. I use it daily. The opening is there.

---

Credits

• TheROCk team - Making gfx1151 usable
• ComfyUI team - Great software
• Lightricks - Open-sourcing LTX-2
• AMD community - Sharing knowledge

---

License

MIT - Use freely, attribution appreciated.

---

Tested: January 2026 | Hardware: GMTEK NUC EVO 2, Strix Halo, 128GB | OS: Ubuntu 25.10

This is what I actually run. Every day. For real work.