VRAM Suite v0.5.0

The first public release: installable from PyPI and as a Windows exe.

Added

• pip install vramsuite. Only the NVIDIA driver is needed: no PyTorch, no CUDA Toolkit.
• pip install "vramsuite[compute]" installs cuBLAS / cuBLASLt from NVIDIA's wheels for doctor --compute-check.
• vramsuite.exe for Windows, built by GitHub Actions. Double-clicking it runs doctor and waits for Enter.
• vramsuite bundle writes an anonymized zip to share: the card, its validation, workflow profiles and a manifest. GPU UUID and the Python path are removed; the home folder, user name and IP addresses are masked.
• vramsuite --version, and python -m vramsuite.

Changed

• pydantic is no longer a dependency; the runtime dependencies are typer and rich.
• README links and images use absolute URLs, so they also work on PyPI.

Install

pip install vramsuite==0.5.0
pip install "vramsuite[compute]==0.5.0"   # + cuBLAS for doctor --compute-check

Or download vramsuite.exe below (Windows, no Python needed) and run vramsuite.exe doctor in a terminal. Double-clicking it runs doctor and waits for Enter.

Requirements

• Windows 10/11 x64 or Linux, an NVIDIA GPU and its driver.
• No Python, PyTorch or CUDA Toolkit needed for the exe.

Limited without the CUDA Toolkit

Everything works with only the NVIDIA driver, except doctor --compute-check (cuBLAS / cuBLASLt FP32 and FP8 checks). It needs the CUDA Toolkit, or pip install "vramsuite[compute]". The exe cannot install extras; with the toolkit installed it finds cuBLAS on its own.

Windows SmartScreen

The exe is not code-signed, so Windows may show "Windows protected your PC" / "Unknown publisher". Click More info → Run anyway, or first check the file against the checksum:

SHA256  611814de681f58249ce9452b72cd3c389c2bf15357ea754735a735d9e376a948  vramsuite.exe

Get-FileHash .\vramsuite.exe -Algorithm SHA256