NVIDIA RTX Spark laptop reviewed: $30K-class Windows AI PC that doubles as a 'token factory'
智东西 · wechat · 2026-10-15
Zhidongxi went hands-on with the ASUS ProArt RTX Spark 16 laptop ($4,100 US pricing, 1.79kg, under 1.3cm thick), built on NVIDIA's N1X superchip: 20-core Grace CPU, 6144-core Blackwell GPU, 64GB unified memory, and 1 PFLOPS FP4 AI compute at 2x the efficiency of an RTX 5070 discrete setup.
Key findings: local Qwen3.8-27B agent benchmark ran in 25 minutes at 23 tokens/s, matching DGX Spark; agents can control drivers and settings via MCP-connected NVIDIA App; paper analysis, invoice parsing, and script review agent tasks worked well. NVIDIA calculates 24/7 local Qwen inference yields 61M tokens/month for $20 in electricity versus $1,200 via cloud APIs — a "personal token factory." Local LoRA fine-tuning via Unsloth and household compute sharing via open-source PAIR are supported. Downsides: only 1.5 hours of battery under heavy AI load, and Arm's gaming ecosystem still lags.
Related event: First RTX Spark Laptops Hit Shelves as ASUS ProArt Debuts in China(3 posts)→
More from Embodied
- China's NeodraDynamics unveils Neo1, a 13-foot industrial mecha with 1,102 lb payload — AIFlow_ML · 2026-10-17
- ALS patient says Neuralink became a lifeline; Musk: it will help millions — elonmusk · 2026-10-17
- Google launches Gemini 4 Argon, a robotics-focused model for physical AI — danielrock · 2026-10-17
- Scoble: OpenAI's Astra is rapidly changing 1X's Neo robot, first units go to developers — Scobleizer · 2026-10-17
- Tesla Roadster uses SpaceX cold-gas thrusters to hit 0-60 mph in 0.55 seconds — elonmusk · 2026-10-17
- OpenAI engineer used his Dot to triage a 3am on-call incident via a Codex thread from his phone — OpenAIDevs · 2026-10-17