NVIDIA GPU

A GPU is optional

An NVIDIA GPU is not required to use Joe's Real Rolls. A compatible NVIDIA GPU can shorten personal-model training time, especially with a larger set of reviewed training images.

What a GPU helps with

The main benefit is faster model configuration and training. Performance varies by computer and training-set size. A faster GPU reduces waiting; it cannot replace accurate labels, balanced examples, stable camera placement, consistent lighting, or patient fresh-roll testing.

Recommended setup

1. Install a current NVIDIA graphics driver and restart Windows if requested.
2. Close other graphics- or AI-heavy applications.
3. Open Training → 2. Train Recognition Model.
4. Select Auto Configure and Start Training.
5. Allow Joe's Real Rolls to choose stable settings for the available hardware.
6. Let training finish before judging the result.

The largest batch number is not automatically the fastest or safest choice. Auto Configure balances speed, memory, and useful learning steps. If a hardware check cannot finish promptly, training can continue with a conservative saved choice.

If acceleration is unavailable

• Confirm Windows recognizes the NVIDIA GPU.
• Update the NVIDIA driver and restart Windows.
• Close other applications using the GPU.
• Run Auto Configure again.
• Continue with CPU training if needed.

GPU and training troubleshooting

The NVIDIA GPU is not being used

1. Confirm Windows Device Manager recognizes the NVIDIA GPU without an error.
2. Install a current NVIDIA graphics driver and restart Windows.
3. Close applications that are heavily using the GPU.
4. Open Train Recognition Model and run Auto Configure and Start Training again.

Joe's Real Rolls can continue with CPU training when acceleration is unavailable. A GPU is not required for the Factory Baseline, camera setup, calibration, or normal recognition.

Auto Configure takes longer than expected

Allow the hardware check to finish and avoid starting overlapping actions. If a hardware check cannot complete promptly, training can continue with a conservative saved choice. Follow the progress shown in the training window.

Training is still slow with a GPU

• Close games, video tools, AI applications, and other graphics-heavy programs.
• Confirm the computer is not in a power-saving mode that limits graphics performance.
• Use the automatically selected settings rather than assuming the largest batch is fastest.
• Remember that larger, carefully reviewed training collections naturally take longer.

Training switches to Full Rebuild

This is a safety response when the current personal model cannot be updated reliably. Let the restarted training run finish. The working model and protected Factory Baseline are kept separate from the new candidate during the process.

Training completed but recognition is worse

This is usually a data or setup issue rather than a GPU issue. Restore a known-good model, review recent images and labels, confirm camera and lighting consistency, and retrain only the affected dice.

The computer has no NVIDIA GPU

Continue using the Factory Baseline and CPU training. Keep sessions focused, let each run finish, and judge the model with fresh physical rolls. Do not buy a GPU solely to begin using Joe's Real Rolls.