Local diagnostic CLI for NVIDIA DGX Spark (GB10). Detects power caps, UMA pressure, thermal risk, CUDA 13/SM_121 wheel mismatches, Docker issues, and validates vLLM/Ollama/llama.cpp/SGLang recipes.
Headless 4K remote desktop for the NVIDIA DGX Spark (GB10): one-command installer for Sunshine + Moonlight low-latency game streaming with NVENC hardware encoding, a software virtual display (no HDMI dummy plug), GDM autologin, and optional Tailscale.
Serve GLM-5.2 469B (REAP-pruned, NVFP4) across 3× NVIDIA DGX Spark with vLLM pipeline parallelism — 256K context, production-ready config and patches
Ubuntu 24.04 + NVIDIA stack setup guide for GB10 / DGX Spark systems
Turn any NVIDIA GPU into a local AI platform. Inference + fine-tuning in your browser. One command to start, automatic clustering.
Terminal system monitor (TUI) that builds its layout from the hardware it actually finds — hybrid CPU clusters, GPUs, RDMA/RoCE NICs, PCIe link state, thermals — plus whole-cluster and fabric-validation views. Born on the NVIDIA DGX Spark (GB10), not limited to it.
Reproducible kit to deploy DeepSeek-V4-Flash-DSpark on a 2× NVIDIA DGX Spark (GB10) cluster: vLLM TP=2 over QSFP 200GbE, NVFP4 KV, DSpark speculative decoding, 1M context, systemd self-heal. Apache-2.0.
Complete setup guide for a 2-node NVIDIA DGX Spark cluster — distributed training, CUDA inference with EXO, NCCL tuning for Grace Blackwell, NVMe-TCP shared storage, and 200 Gb/s direct fabric networking.
GLM-5.2 QuantTrio (Unpruned ) on 4xDGX Spark (GB10): 327K-655K context or up to 5 concurrent agents, one receipe. TP4 + DCP + MTP
DeepSeek-V4-Flash DSpark speculative decoding on 2x DGX Spark (GB10/sm_121), 1M context — our fp8 recipe + independent reproduction and cross-build benchmarks of the NVFP4-KV build. Honest, apples-to-apples measurements.
Practical AI homelab setup guides for GB10, Mac Studio Ultra, RoCE/RDMA, MikroTik switching, NCCL, and heterogeneous workload experiments.
Qwen3.8-Flash-Next as Cogni-Brain on NVIDIA DGX Spark (GB10): HashK GPU PLE + SGLang NEXTN, 36.8 tok/s code, 100/100 tool-eval, 262K context.
Real, reproducible LLM inference benchmarks on the NVIDIA DGX Spark — parallel agent sessions, tok/s and latency under load, with an interactive dashboard and one-command recipes
Turn a DGX Spark GPU cluster into a self-running team of specialized AI agents — cluster control, role-specialized workers, and an autonomous review gate for Hermes
NVML unified memory shim for NVIDIA DGX Spark Grace Blackwell GB10 - enables MAX Engine, PyTorch, and GPU monitoring
Pre-built PyTorch wheels and build scripts for NVIDIA DGX Spark (GB10, sm_121, Blackwell, CUDA 13.0, ARM64)
Build paddlepaddle-gpu from source for NVIDIA DGX Spark and GB10 (Grace Blackwell, aarch64, CUDA 13, sm_121). Prebuilt wheel plus the exact patches upstream is missing.
Reproducible MiniMax-H3 benchmarks on NVIDIA GB300: SGLang scaling, host eBPF CUDA tracing, Grace-GPU PMU traffic, energy, and Nsight comparison.
Measured vLLM serving for 1–2 NVIDIA DGX Spark (GB10): TP=2 RoCE flagship, validation ledger, stranger-ready Path A/B onboarding.
Field notes from optimizing LLM inference on an NVIDIA DGX Spark. Everything measured on the machine.
To associate your repository with the grace-blackwell topic, visit your repo's landing page and select "manage topics."