NVIDIA pushes local AI at IFA 2026: 1.9x faster inference, PAIR, RTX Spark PCs

AWS ML Blog · rss · 2026-09-04

NVIDIA, Microsoft and partners announced a slate of local AI updates at IFA 2026:

• Faster inference: llama.cpp gains up to 1.9x throughput on RTX 5090; vLLM is 1.2x on RTX PRO 6000 and up to 1.4x on dual DGX Spark clusters, available via LM Studio and Ollama.
• NVIDIA PAIR: a free, open-source Personal AI Router that discovers compatible PCs on a local network and distributes inference across idle machines.
• One-click local agents: Hermes Agent, OpenClaw (380K+ GitHub stars) and Perplexity Portable Computer will support simplified local model setup on Windows; Portable Computer runs workflows locally and selectively escalates to 15+ frontier cloud models.
• RTX Spark: Lenovo and Acer Windows PCs arrive in October.
• New locally runnable models: Nemotron 3.5 Lightning (30B), GLM-5.3-Flash, Qwen3.8-Flash-Next/27B, LTX 2.5 video, MiniMax-H3 plus FastH3 (4-step distilled, 7x faster), Meta Muse Glimmer (30B), and DeepSeek v4 Flash (284B MoE, 13B active, runs on 2x DGX Spark).

Original post →

More from Embodied

• Perceptron's Isaac 0.5 robot folds t-shirts, ships open weights for cross-embodiment repro — iamrobotbear · 2026-09-04
• Lab lessons from Anthropic MHS: keep fast control out of the model — Empty-Abalone-2952 · 2026-09-04
• Build or OEM? Nvidia Fabless Lesson for Humanoid Robot Makers — chris_j_paxton · 2026-09-04
• Sentdex shows general multimodal LLMs can drive robots with zero training — Sentdex · 2026-09-04
• "Atlas <> Palantir" teaser surfaces, pointing to a September 10, 2026 reveal — eliano · 2026-09-04
• Ultra's hybrid robotic-arm plus semi-humanoid strategy targets rapid 3PL field deployment — chris_j_paxton · 2026-09-04

Embodied channel →