AI BriefingNVIDIA Showcases Local AI Hardware; Recursive Superintelligence Releases Flint for RTX Spark
·2026.10.21 19:50
Key point
NVIDIA showcases its local AI hardware lineup at GTC Berlin, while Recursive Superintelligence releases the Recursive Flint model optimized for RTX Spark.
Details
- NVIDIA showcases its full local AI hardware lineup at GTC Berlin, including RTX Spark for Windows PCs, DGX Spark for Linux developers, and DGX Station for enterprise teams.
- Recursive Superintelligence releases Recursive Flint, a 40-billion-parameter mixture-of-experts model, which achieves over 220 tokens per second on 64GB RTX Spark and tops agentic coding benchmarks like SWE-Bench Pro v2.
- DGX Spark 64GB systems begin rolling out on October 23, supporting clusters of up to eight units for 1TB of unified memory to run frontier-class models like GLM 5.3.
- DGX Station provides 748GB of coherent memory and up to 20 petaflops of AI performance, with a Windows version launching in December to support manufacturing apps and agent harnesses.
- Day-zero support is added for new models including Nemotron Bolt E4B 4B, GLM, DeepSeek, Qwen, MiniMax, Liquid AI’s LFM2.5-24B-A2B, and FLUX.3 Klein.
- Inference optimizations for llama.cpp and vLLM accelerate agentic tasks by up to 1.4x on RTX PRO 6000 Blackwell and 1.3x on DGX Spark.
- Three new open source reference workflows for robotics on DGX Station enable post-training, simulation, and evaluation using GPT-6 Astra, NVIDIA Isaac Lab, and Cosmos.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
DeepSeek open-sources DeepInfer inference framework for V4.1-Flash