NVIDIA DGX Spark (GB10 Grace Blackwell Superchip)
Palm-sized personal AI supercomputer powered by the NVIDIA GB10 Grace Blackwell Superchip with 128GB coherent unified LPDDR5X memory and ConnectX-7 200GbE clustering.
Key Engineering Highlights
- 128GB of coherent unified system memory runs 70B FP8 or 200B FP4 models locally on a single desk box
- Built-in ConnectX-7 NIC links two units to run Llama-4-405B at low latency
Full Hardware Specifications
| Processor / Edge SoC | NVIDIA GB10 (20-Core Arm Grace + Blackwell GPU) |
|---|---|
| AI Model Stack | Runs Up to 200B Parameter Models (405B Clustered) |
| Display & Optics | HDMI 2.1a + 4x USB4 Type-C |
| Camera & Vision | Network / USB-C CSI |
| Measured Weight | 1,200g (Book-Sized Chassis) |
| Battery & Runtime | 170W USB-C GaN Desktop Power Brick |
| Connectivity & I/O | NVIDIA ConnectX-7 SmartNIC + Wi-Fi 7 |
| Sensors & Acoustics | 1,000 TOPS (1 PFLOP) Sparse FP4 Tensor Cores |
⚖️ Side-by-Side Comparison: NVIDIA DGX Spark (GB10 Grace Blackwell Superchip) vs. Edge Compute & DevKits Peers
Direct specification comparison against leading competing devices in the Edge Compute & DevKits category.
| Specification | NVIDIA DGX Spark (GB10 Grace Blackwell Superchip) (Current) | NVIDIA Jetson Orin NX Super Edge AI DevKit → | Xiaozhi AI Voice Cube Pro (ESP32-S3 Open DevKit) → | Apple Mac mini (M5 Pro 64GB) Private AI Node → |
|---|---|---|---|---|
| Brand | NVIDIA | NVIDIA | Open Hardware Community | Apple |
| Retail Price | $2,999 | $499 | $12 | $1,799 |
| Lab Rating | ||||
| Edge SoC | NVIDIA GB10 (20-Core Arm Grace + Blackwell GPU) | NVIDIA Ampere GPU (1024 CUDA + 32 Tensor Cores) | Espressif ESP32-S3R8 (Dual-Core LX7 @ 240MHz + 8MB PSRAM) | Apple M5 Pro (14c CPU + 20c GPU + 16c Neural Engine) |
| AI Model Stack | Runs Up to 200B Parameter Models (405B Clustered) | Runs Llama-3.2 8B / Qwen2.5-VL 7B / YOLO11 Locally | Edge ESP-SR Wake + Realtime Cloud Voice LLMs | MLX Accelerated Qwen3-32B / Llama-3.3-70B |
| Display / Optics | HDMI 2.1a + 4x USB4 Type-C | HDMI 2.1 / DP 1.4 (Up to 8K) | 1.28-inch Circular IPS LCD (240×240) / 0.96" OLED | Supports up to three 6K displays |
| Camera | Network / USB-C CSI | Dual MIPI CSI-2 Camera Lanes + 4× USB 3.2 | Optional OV2640 / GC2145 Camera Module | External USB-C / Studio Display Camera |
| Weight | 1,200g (Book-Sized Chassis) | 450g (including active heatsink & carrier board) | 62g (with 3D-printed enclosure & speaker cavity) | 730g |
| Battery | 170W USB-C GaN Desktop Power Brick | 19V DC Input (15W–40W configurable TDP) | 800mAh LiPo (5h continuous dialogue) | Internal 155W PSU (4W idle draw) |
📚 Related Engineering Teardowns & Reviews
Browse All 83 Articles →Framework Desktop (Ryzen AI Max+ 395 128GB) Local LLM Benchmark: AMD Strix Halo vs NVIDIA DGX Spark GB10 vs Mac mini M5 Pro
RECIWO Labs benchmarked the $1,999 4.5L Framework Desktop powered by AMD's 16-core Ryzen AI Max+ 395 ('Strix Halo') and 128GB unified LPDDR5X-8000 memory, comparing ROCm 6.4 token generation and prompt prefill speeds against the $2,999 NVIDIA DGX Spark (GB10) and $1,799 Apple Mac mini M5 Pro.
NVIDIA DGX Spark (GB10 128GB) Desk Supercomputer Review: Running 200B FP4 Local LLMs at 140W
RECIWO Labs tested the $2,999 NVIDIA DGX Spark (GB10 Grace Blackwell Superchip, 128GB LPDDR5X) against the 64GB Mac mini M5 Pro and dual RTX 5090 towers across 70B FP8 and 200B NVFP4 models.
INT4 AWQ vs. GPTQ vs. FP8 Quantization on Edge NPUs: Measuring Perplexity & Tool-Call Degradation
RECIWO Labs tested 7B, 32B, and 70B models across FP16, FP8, INT4-AWQ, and INT4-GPTQ on edge NPUs to measure exact coding and JSON tool-calling accuracy loss.
Speculative Decoding on Edge Hardware (EAGLE-2 & Draft Models): 2.4x Faster Local LLM Generation
How pairing a 0.6B draft head or EAGLE-2 feature predictor with a 32B/70B target model turns idle edge NPU arithmetic capacity into 2.4x faster token streaming.
acoustics & Thermals of Compact Local AI Nodes: Why Phase-Change Vapor Chambers Beat AIO Liquid Loops
We measured 24/7 pump failure rates and 1-meter dBA noise floor across compact local AI workstations running continuous night-long RAG indexing and LoRA fine-tuning.
