cgpu-info 0.3.0

Quick CUDA/GPU status summary for ML engineers

cgpu

Quick CUDA/GPU status summary for ML engineers. One import, one call, all the info you need.

Installation

uv pip install cgpu-info

Or with pip:

pip install cgpu-info

Usage

Python

from cgpu import cgpu

device = cgpu()
# Now use `device` in your code
model.to(device)

Progress bar with GPU metrics

from cgpu import ctqdm

# Drop-in replacement for tqdm - shows GPU temp, VRAM, and utilization
for batch in ctqdm(dataloader, desc="Training"):
    ...

Output:

Training:  50%|█████     | 50/100 [00:05<00:05] 42°C | 8.2/24.0GB | GPU: 87%

Install nvidia-ml-py for temperature and utilization. Without it, VRAM usage is still shown via torch.

CLI

# Show GPU status
cgpu

# Show version
cgpu --version

That's it! You'll see a colorful summary like:

═══════════════════════════════════════
          GPU Status Summary
═══════════════════════════════════════
✓ CUDA Available
  Device: cuda
  GPU Count: 1
  [0] NVIDIA GeForce RTX 4090
      VRAM: 24.0 GB
      Allocated: 0.00 GB
      Reserved: 0.00 GB
      Temp: 42°C
      GPU Util: 0%
      Mem Util: 0%
  CUDA Version: 12.1
  cuDNN Version: 8902
  PyTorch: 2.1.0
═══════════════════════════════════════

What it shows

• CUDA availability status
• Device string (cuda or cpu)
• GPU name and count
• VRAM total and usage
• Temperature (color-coded: green < 50°C, yellow < 70°C, red >= 70°C)
• GPU/Memory utilization
• CUDA, cuDNN, and PyTorch versions

Installing PyTorch with specific CUDA version

Use the built-in install helper to get PyTorch with your desired CUDA version:

# Install torch with CUDA 12.1
cgpu install --cuda 12.1

# Install torch with CUDA 12.4
cgpu install --cuda 12.4

# Install torch with CUDA 11.8
cgpu install --cuda 11.8

# Install CPU-only torch
cgpu install --cuda cpu

Optional dependencies

If you want to install additional ML packages along with cgpu:

# Full ML stack: torch, torchvision, torchaudio, numpy, pandas, matplotlib, seaborn, scikit-learn
uv pip install cgpu-info[full]

# Just PyTorch stack
uv pip install cgpu-info[torch]

# Just data science packages (no torch)
uv pip install cgpu-info[science]

License

MIT