Hardware Settings

Configuring hardware acceleration, GPU settings, and the Python runtime environment

The Hardware Settings screen configures how Studio Desktop uses your computer's hardware during training and inference. This includes selecting between CPU and GPU processing, setting the numerical precision, and managing the Python runtime environment.

Hardware Configuration

Session Provider

Choose the compute device for training:

Provider	Description
GPU	Uses your NVIDIA GPU via CUDA. Dramatically faster for training.
CPU	Uses the processor only. Significantly slower but does not require a GPU.

When GPU is selected, the detected GPU name and available VRAM are displayed below the selector. If no GPU is detected, the display shows "No GPU detected" and you should verify your NVIDIA drivers.

Precision

Select the numerical precision used during training:

Precision	Description	When to Use
16-mixed	Mixed precision with 16-bit floating point	Default — fastest training with minimal accuracy loss
bf16-mixed	Mixed precision with bfloat16	For newer GPUs (Ampere architecture or later) that support bfloat16 natively
32-bit	Full 32-bit floating point	Maximum precision — use if you experience training instability with mixed precision

Mixed precision (16-mixed) is recommended for most users. It uses roughly half the GPU memory compared to 32-bit and trains faster, with negligible impact on model accuracy.

Number of Devices

Set the number of GPUs to use for training. For most setups, this is 1. Multi-GPU configurations can accelerate training on systems with multiple NVIDIA GPUs.

Saving

Click Save after changing any settings. A confirmation dialog appears if you navigate away with unsaved changes. All values are stored in the project's config.yaml.

Runtime Environment

The runtime environment section manages the self-contained Python installation that Studio Desktop uses for all ML operations.

Checking Runtime Status

When you open Hardware Settings, the application automatically checks the runtime status. There are three possible states:

State	Meaning	Action
Installed and up-to-date	The correct Python version is installed	No action needed; the version number is displayed
Outdated	A newer Python version is required	Click Update Runtime
Not installed	No Python runtime found	Click Install Runtime

Installing or Updating the Runtime

1. Click Install Runtime (or Update Runtime).

2. The application downloads a portable Python distribution (approximately 1–2 GB).

3. Progress is displayed with a progress bar and percentage.

4. Once extraction completes, the status updates to "Installed."

An internet connection is required for the initial download. The runtime is installed to the application's user data folder and does not interfere with any existing Python installation on your system.

Checking GPU Availability

Click Check GPU to run a diagnostic that verifies your GPU is accessible to the Python runtime. This spawns a small Python script that:

1. Checks if CUDA is available.

2. Reports the GPU name, VRAM, and CUDA version.

3. Confirms that PyTorch can access the GPU.

The Check GPU button is only available after the Python runtime is installed.

Troubleshooting

Issue	Possible Cause	Solution
"No GPU detected"	NVIDIA drivers not installed or outdated	Install the latest NVIDIA drivers from nvidia.com
GPU detected but training uses CPU	Session provider is set to CPU	Switch the session provider to GPU and save
Runtime download fails	No internet connection or firewall blocking	Ensure internet access; check proxy/firewall settings
Training instability with 16-mixed	GPU does not fully support mixed precision	Switch to 32-bit precision
Multiple GPUs not recognized	Drivers or CUDA toolkit not configured for multi-GPU	Verify NVIDIA driver configuration; typically only one GPU is needed