title: README
emoji: 📈
colorFrom: purple
colorTo: green
sdk: static
pinned: false
Enverge.ai
Spark lets you rent NVIDIA DGX Spark devices — each with 128GB of unified memory — to train, fine-tune, and run large AI models. Zero setup, no infrastructure to manage: launch a DGX Spark in seconds and start building.
📄 See the official NVIDIA DGX Spark spec sheet for full hardware details.
👉 Rent a DGX Spark at spark.enverge.ai
Recent blog posts
From spark.enverge.ai/blog:
How fast is the DGX Spark, really? Prefill vs. decode, and the 273 GB/s wall — Why DGX Spark decode tops out around 3 tok/s on dense 70B models — and why prefill, MoE models, and batched serving tell a very different story.
The Cheapest Way to Run a 70B Model Locally in 2026 — The cheapest way to run a 70B model locally, compared: DGX Spark, GB10 clones, Mac Studio, RTX 5090, and cloud rental — with specs, prices, and break-even math.
How (and Why) to Quantize LLMs on NVIDIA DGX Spark — Quantize LLMs on NVIDIA DGX Spark using NVFP4, FP8, and GGUF. Step-by-step calibration, evaluation, and tradeoffs for Llama 3.1 70B — under $2 of compute.
Running Research Experiments on DGX Spark: Why Smaller VRAM Can Be Cheaper for Iterative AI — Why H100s are overkill for iterative research — and how DGX Spark at $0.65/hr lets you run 5–8x more experiment variants for the same budget.
Run AI Agents Locally: OpenClaw, Local LLMs, and Why the Cloud Should Be Yours — Why building AI agents on API calls is expensive and insecure — and how running OpenClaw with local LLMs on Spark Cloud keeps your data private while cutting costs by half.