Razer, Blade Stealth 13, 13.3″, Gaming Laptop Black RZ09-03102E22 | Node Titan | AMD RX 6800

/ all-scribbly / By

After an extended battle with a mac-mini a gpu and a container – I finall retreated back toward someting that should work better out of the box and woudl not require me selling it all – and reuiliding it all again.

Razer, Blade Stealth 13, 13.3″, Gaming Laptop Black RZ09-03102E22 – windows-only with your RX 6800 in the Node Titan. Here’s a clean, don’t-go-in-circles checklist that gets you from fresh install → AI-ready on Windows 11 Pro Preferred

1) Plug-and-play basics (one time)

  1. Update Windows 11 fully. Reboot.

  2. Install Intel Thunderbolt Control Center from Microsoft Store (so TB3 trusts the eGPU when you plug it in).

  3. Connect the Node Titan directly to a TB3 port (no hubs), then power the enclosure first, then attach TB3. You should see “Approve” in Thunderbolt Control Center the first time.

  4. Install AMD Software: Adrenalin Edition (WHQL) for Radeon RX 6800. Use “Factory Reset / Clean install” if prompted.

    • If you ever need a prior stable build, AMD keeps a “Previous Drivers” list.

  5. Optional but nice: OWC/AKiTiO docs confirm Node Titan is TB3 eGPU-ready; it’s routinely used with RX 6800/6800 XT.

2) Make Windows prefer the eGPU

  • Windows Settings → System → Display → Graphics → “Default graphics settings”: enable Hardware-accelerated GPU scheduling.
  • Graphics → Browse and add the apps you care about (e.g., python.exe, your IDE, game executables) → set High performance (that picks the RX 6800).
  • Leave Intel iGPU and the laptop’s GTX 1650 enabled; Windows will route displays, but your apps will hit the RX 6800 when you assign them.

  1. After Windows Update, install Razer/Intel platform bits before AMD:

    • From Razer’s page for RZ09-03102: Intel Chipset / ME, Intel iGPU, Realtek Audio, Intel Wi-Fi & BT.
    • Then do AMD Adrenalin (RX 6800) clean install.
    • Reason: ensures PCIe/Thunderbolt power mgmt + audio stack are clean before the big GPU driver.

    1. External display = free fps.

      Plug your monitor into the Node Titan’s HDMI/DP when you care about performance. Driving the internal panel over TB3 costs ~10–20% in many apps.

  1. Windows graphics prefs (you listed) + one extra:

    • Settings → System → Power → Power mode: Best performance (on AC).
    • Control Panel → Power Options → Change plan → Advanced → PCI Express → Link State Power Management = Off (helps eGPU stability under load).

  2. Thunderbolt app:

    If the Microsoft Store “Thunderbolt Control Center” refuses to install, grab the Intel Thunderbolt DCH driver first (then the Store app will open and show the “Approve” prompt). Some builds need that driver layer.

  3. Optional stability when swapping GPUs often:

    If you’ve had old driver remnants, run AMD “Factory Reset” (you noted) or DDU (safe mode) once, then stick to Adrenalin WHQL going forward.

setup-ai-env.ps1

# Elevate privileges if not running as adminif (-not ([Security.Principal.WindowsPrincipal] [Security.Principal.WindowsIdentity]::GetCurrent()).IsInRole([Security.Principal.WindowsBuiltInRole] "Administrator")){    Start-Process powershell -Verb runAs -ArgumentList "-File `"$PSCommandPath`""    exit}Write-Host "=== Installing Python 3.11 via Winget (if missing) ==="winget install --id Python.Python.3.11 --source winget -e --accept-package-agreements --accept-source-agreements# Ensure Python is on PATH$python = (Get-Command python.exe -ErrorAction SilentlyContinue)if (-not $python) {    $env:Path += ";$($env:LocalAppData)\\Programs\\Python\\Python311"}Write-Host "=== Creating venv at C:\\AIenv ==="New-Item -ItemType Directory -Force -Path "C:\\AIenv" | Out-Nullcd C:\\AIenvpython -m venv venv.\\venv\\Scripts\\activateWrite-Host "=== Upgrading pip and installing PyTorch DirectML ==="python -m pip install --upgrade pippip install torch-directmlWrite-Host "=== Running verification test ==="python - <<'PYCODE'import torch_directml as dml, torchdevice = dml.device(dml.default_device())x = torch.randn(1024, 1024, device=device)print("✅ DirectML active on:", device)print("Tensor multiply OK:", (x @ x).sum().item())PYCODEWrite-Host "=== Setting Python to High Performance GPU profile ==="Add-ItemProperty -Path "HKCU:\\Software\\Microsoft\\DirectX\\UserGpuPreferences" `    -Name "C:\\AIenv\\venv\\Scripts\\python.exe" `    -Value "GpuPreference=2;" -ForceWrite-Host "`n✅  Setup complete!  Run 'C:\\AIenv\\venv\\Scripts\\activate' in PowerShell to start your AI environment."
PowerShell

🧩 What it does

Step Action
1 Installs Python 3.11 (winget)
2 Creates C:\\AIenv\\venv virtual environment
3 Installs torch-directml for AMD GPUs
4 Runs a quick tensor test
5 Writes a registry key that makes python.exe use your RX 6800 (High Performance) automatically

Run it once:

Right-click → Run with PowerShell (Admin)

After it finishes, reopen PowerShell and:

C:\\AIenv\\venv\\Scripts\\activatepython>>> import torch_directml

You’ll see confirmation that DirectML is active on your Radeon RX 6800 — no manual driver juggling needed.

3) AI stack on Windows (two good paths)

A) Fast path (works onany** DX12 GPU):

PyTorch + DirectML

This is the most friction-free way to get PyTorch using your RX 6800 on Windows.

py -3.11 -m venv venvvenv\\Scripts\\activatepip install --upgrade pippip install torch-directml
  • This is Microsoft’s maintained package; latest guide here.

Smoke test:

import torch, torch_directml as dmldevice = dml.device(dml.default_device())x = torch.randn(2048, 2048, device=device); y = x @ xprint("DirectML device:", device, "OK")

(If that prints a DirectML device, you’re accelerating on the RX 6800.)

B) (let’s try if hit bottlenecks) Higher-performance path:ROCm on Windows (RX 6800 supported)

AMD has rolled out ROCm for Windows in 2H-2025; RX 6800 (RDNA2, gfx1030) appears on the Windows-supported GPUs list for the HIP SDK/runtime.

Install (Preview cadence evolves—follow AMD’s notes):

winget install AMD.ROCm

Then a ROCm build of PyTorch (exact wheel may change as AMD updates releases):

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.1

AMD’s current release notes call out Windows PyTorch preview availability; check those if a newer wheel is posted.

Test (ROCm):

import torchprint(torch.__version__)print("HIP available:", torch.version.hip is not None)

Tip: If you hit any early-preview bumps, fall back to DirectML for instant productivity while ROCm matures on Windows.

4) Optional apps you’ll probably like

  • Ollama (Windows, AMD) – easy local LLMs; AMD acceleration is supported (preview).
  • Stable Diffusion UI builds also run via DirectML on AMD; if a PyTorch wheel mismatch pops up, check the app’s readme for DirectML steps.

5) Quality-of-life tuning

  • Adrenalin → Performance → Tuning: apply a gentle fan curve and a small power limit +10% to avoid throttling during long runs.
  • Keep the laptop on AC power when using the eGPU.
  • If the enclosure disconnects after sleep, disable “USB selective suspend” in your current power plan.

6) Driver & firmware links you may want handy

  • Razer Blade Stealth 13 (RZ09-03102) — Drivers/BIOS page (chipset, ME, audio, Wi-Fi, etc.).
  • AMD Adrenalin for RX 6800 (WHQL / Optional).
  • DirectML PyTorch docs (quick install).
  • ROCm on Windows announcement / GPU list (for context & support table).

Here’s a clean, safe PowerShell script you can copy, save as setup-ai-env.ps1, and run once after Windows is updated and your RX 6800 drivers are installed.

It creates a self-contained Python 3.11 environment, installs PyTorch DirectML, and pins that Python process to your AMD GPU in Windows 11’s graphics settings.

⚙️

Related Posts