Nvidia undervolting companion on Linux. Can automatically scan for GPU voltage-frequency curves from the most efficient to the most performant, and apply them adaptively based on the pre frame-gen FPS. MSI Afterburner profile imports and LACT profile exports.

Local, portable GUI for Qwen3-TTS. Optimized for NVIDIA RTX 50 Series (CUDA 12.8). One-click install.

A hands-on guide for AI builders: make your own RTX 4090D/5090 GPU server that’s fast and efficient.

A local Gradio Web-UI for the VIBE model (https://huggingface.co/iitolstykh/VIBE-Image-Edit), supporting advanced features...

GPU-tuned Docker images for LLM inference on consumer hardware. Auto-detects your GPU, downloads the model, serves an OpenAI-compatible API.

A lightweight, high‑performance local interface for **FLUX.2 Klein (4B)** by Black Forest Labs.

Optimized vLLM deployment for NVIDIA Blackwell (RTX 5090) on Linux Kernel 6.14. Resolves SM_120 kernel incompatibilities, P2P deadlocks, and memory fragmentation for high-performance LLM inference.

⚡ A lightning-fast local Gradio GUI for Z-Image-Turbo (Windows). Optimized for NVIDIA RTX 5090 (Blackwell) & CUDA 12.8.

Small app that checks api.store.nvidia.com for 50XX stock

Environment setup and build fixes for 3D Gaussian Splatting on RTX 5090 (Windows, CUDA 12.8) ; See setup instructions in README_rtx5090setup.md

A lightweight, high-performance local interface for **FLUX.2 Klein (9B)** by Black Forest Labs. This project is specifically optimized for **NVIDIA RTX 5090** GPUs, utilizing PyTorch Nightly (CUDA 13.0) and `bfloat16` precision for maximum speed.

RTX 5090 Cinema AI Workstation - 325 apps installed in 27 minutes via Claude Code + Pinokio

Enable utility for MSI Display Device when the onboard display turns black because the Windows device is disabled.

Restart utility for MSI Lightning Service when the onboard display stops updating hardware monitoring data on MSI GeForce RTX 5090 LIGHTNING Z.

Self-hosted sovereign AI stack on one RTX 5090 — vLLM + Nemotron NVFP4 (275 tok/s, 180K ctx), routing-enforced privacy tiers, full decision log

Local AI video production on RTX 5090 — FLUX.2/Wan 2.2/HunyuanVideo generation, VOID object removal, SAM3+VACE editing, RIFE+SeedVR2 4K60 finishing

Experimental Windows MSLK build for NVIDIA Blackwell (SM120) GPUs. Includes a validated Triton Split-K adjustment for head_dim=512 workloads used by xFormers.

A fully automated PowerShell script to compile PyTorch from source with CUDA 12.1 support for NVIDIA RTX 50-series GPUs, optimized for Windows 11.

Windows RTX 50 RealESRGAN anime video upscaling pipeline, TensorRT/vs-mlrt benchmarks, and SM120 experiments