rtx-spark

Serve the home! Inference stack for your Nvidia DGX / RTX Spark (128GB ram) aka the Grace Blackwell AI supercomputer on your desk. Mostly vLLM based for now and single-spark. For the not-so-rich people. Updated once a month. Hobby project

Linux on the ASUS ProArt P16 H7607BA (NVIDIA N1X / RTX Spark, arm64): fixes battery 0%, no sound, keyboard/touchpad, keyboard backlight, Fn keys, power/fan modes, charge limit, sensors, haptic touchpad and snap GPU acceleration on Ubuntu 26.04.

Independent community list for NVIDIA RTX Spark (N1X): native Linux, drivers, kernels, inference and benchmarks.

Linux hardware enablement for ASUS ProArt P16 H7607BA with NVIDIA RTX Spark N1X: Omarchy ARM64 installation, kernel patches, ACPI fixes, NVIDIA setup and power/thermal monitoring.

vLLm recipe for running Qwen3.8-Flash-Next on a single DGX Spark and RTX Spark family + DFlash2 support

Run DeepSeek-V4.1-Flash-1017 (ISTA 1.75/2.25bpw) locally on NVIDIA RTX Spark N1X 128GB laptops: Windows ARM64 CUDA engine, Strata desktop and OpenAI-compatible API for DeepSeek Harness (DSH / DeepSeekHarness). 微软 N1X 128G/128GB 本地私有化部署 DeepSeek1017；社区预览，含安装包与已有记录。

Experimental Linux driver and cuDLA backend for the NVIDIA N1X NPU/DLA on ASUS ProArt P16 RTX Spark, with verified inference, module reload and 3x super-resolution.

Fused FP8 Triton matmul for vLLM on NVIDIA RTX Spark / GB10 (sm_121), native Windows. 13% faster than WSL2 with Gemma 4 12B.

RTX Spark N1X / MSI OLED flicker when typing on Windows ARM64: visible Ball or Canvas markedly eased flashes on one tested laptop. English and Chinese apps; temporary workaround.

Mayfly: an autonomous drone swarm for indoor search and rescue. Many small, affordable drones explore a damaged building in parallel and fuse what they see into one shared 3D map, feeding a rescue dashboard with live feeds, survivor locations, estimated vitals, hazards and routes. CalHacks 13.

Blackfrost-AI CYBER-FROST-3.8 (Qwen3.8) on the NVIDIA RTX Spark, native Windows 11 ARM64: prebuilt engine, one-shot setup for you or your agent, performance data

Self-hosted control panel that turns an NVIDIA RTX Spark (N1X) laptop into a 24/7 local-AI server: monitoring, Ollama / vLLM / TensorFold, Docker, Tailscale, JupyterLab, VS Code, opencode — all behind one port.

Local LLMs with vLLM on the NVIDIA RTX Spark N1X (DGX Spark / GB10 family): unified-memory fixes, speed and quality benchmarks of Qwen3.8 27B and Flash-Next, coding-harness comparison

Experimental Android compatibility tooling for NVIDIA RTX Spark (N1X) on Windows ARM64: WSA Android 13 and QEMU/Redroid backends.