NVIDIA Expands RTX Spark, DGX Spark and DGX Station Lineup at GTC Berlin
NVIDIA announced new hardware availability, software updates and day-zero model support across its RTX Spark, DGX Spark and DGX Station local AI systems at GTC Berlin.
Respuesta rápida
What did NVIDIA announce for its local AI hardware lineup at GTC Berlin?
NVIDIA announced updates across its local AI hardware at GTC Berlin: DGX Spark 64GB systems shipping October 23, DGX Station gaining new management integrations and a December Windows version, NVIDIA Sync adding cluster and model-launcher features, and day-zero support for new models including Nemotron Bolt, GLM, DeepSeek, Qwen and FLUX.3.
Claves
- NVIDIA says DGX Spark 64GB systems, able to run models like Nemotron 3.5 Bolt 74B 24GB, Qwen3.8 27B and Gemma 4 26B-A4B, begin rolling out from partners on October 23.
- NVIDIA says DGX Spark now supports manual clustering of up to eight units, delivering 1TB of unified memory for running frontier-class models such as GLM 5.3.
- NVIDIA says DGX Station for Windows will launch in December with support for manufacturing applications from partners including Dassault Systèmes, Siemens and Synopsys, plus agent harnesses from Hermes Agent, Perplexity and Aible.
- NVIDIA says frontier AI lab Recursive Superintelligence released Recursive Flint, a 40-billion-parameter mixture-of-experts coding model with 2.5 billion active parameters, available as open weights under Apache 2.0 for RTX Spark and DGX Spark hardware.
- NVIDIA says it added day-zero support for new models including Nemotron Bolt E4B 4B, GLM, DeepSeek, Qwen, MiniMax, Liquid AI's LFM2.5-24B-A2B and the FLUX.3 Klein diffusion model, alongside inference speedups for llama.cpp and vLLM.
NVIDIA Updates Local AI Hardware Lineup at GTC Berlin
NVIDIA used GTC Berlin to detail a round of hardware availability dates, software updates and new model support across its three local AI systems: RTX Spark, DGX Spark and DGX Station. The company says RTX Spark runs personal agents on Windows PCs, DGX Spark targets developers building on Linux, and DGX Station serves enterprise teams running larger workloads, with a Windows version of DGX Station arriving in December.
DGX Spark Gets Clustering and New Software
NVIDIA says DGX Spark 64GB systems, capable of running models including Nemotron 3.5 Bolt 74B 24GB, Qwen3.8 27B and Gemma 4 26B-A4B, begin shipping from partners on October 23. The company says DGX Spark now supports manual clustering of up to eight units using built-in ConnectX-7 networking and a new NVIDIA Sync Cluster Assistant, providing 1TB of unified memory for running models such as GLM 5.3 across concurrent development sessions.
NVIDIA says an NVIDIA Sync Model Launcher is coming later this month to simplify downloading models and containers for a single DGX Spark or a cluster, and that an October software update adds a Wi-Fi Assistant for switching networks in appliance mode along with updated DGX Spark Skills for troubleshooting guidance.
DGX Station Adds Enterprise Integrations
For larger teams, NVIDIA says DGX Station offers 748GB of coherent memory and up to 20 petaflops of AI performance, and can serve more than 30 users at speeds the company compares to cloud models. NVIDIA says it added seven new management integrations for DGX Station, including with Absolute Security, Splunk by Cisco, Microsoft Intune and Progress Software Chef, to help IT teams manage approvals and deployment.
The Windows version of DGX Station, expected from manufacturer partners in December, will include support for manufacturing applications from companies including Dassault Systèmes, Siemens, Synopsys and MathWorks, plus agent tools from Hermes Agent, Perplexity and Aible, according to NVIDIA. The company also says NVIDIA Cloud Functions is coming in early access to the Linux-based DGX Station, letting teams register local model endpoints through a shared, OpenAI-compatible catalog. NVIDIA additionally released three open source reference workflows for robotics development on DGX Station.
RTX Spark and a New Coding Model
NVIDIA says RTX Spark, which launched October 16, runs personal agents with up to 128GB of unified memory and 1 petaflop of AI compute. The company says a new desktop agent app called Underdog is now available for RTX Spark, running models entirely on-device through its own inference engine.
Separately, NVIDIA says frontier AI lab Recursive Superintelligence released Recursive Flint, a local coding model with 40 billion total parameters and 2.5 billion active parameters, available as open weights under the Apache 2.0 license for RTX Spark and DGX Spark hardware as well as GeForce RTX and RTX PRO GPUs.
New Model Support and Inference Speedups
NVIDIA says it is adding day-zero support for new models including Nemotron Bolt E4B 4B, GLM, DeepSeek, Qwen, MiniMax, Liquid AI's LFM2.5-24B-A2B and the FLUX.3 Klein diffusion model from Black Forest Labs. The company says it also released a new collection of quantized model checkpoints optimized for memory efficiency on local hardware.
NVIDIA says it continues to work with the llama.cpp and vLLM open source projects, adding speculative decoding support that it says makes agentic tasks with Qwen 3.8 Flash Next up to 1.4 times faster on RTX PRO 6000 Blackwell GPUs, and new kernels that speed up Qwen3.8 27B tasks on DGX Spark and GLM 5.3 Flash throughput on DGX Station.
Source: NVIDIA Blog, "Bringing the Spark to Europe: Local Intelligence Powers Next Wave of Agents," October 21, 2026.
Preguntas frecuentes
- When do new DGX Spark 64GB systems become available?
- NVIDIA says DGX Spark 64GB systems start rolling out from partners on Friday, October 23.
- What is Recursive Flint?
- NVIDIA says Recursive Flint is a new local coding model from Recursive Superintelligence, a 40-billion-parameter mixture-of-experts model with 2.5 billion active parameters, released as open weights for RTX Spark and DGX Spark hardware.
- When is DGX Station for Windows launching?
- NVIDIA says DGX Station for Windows is expected to launch in December from system manufacturer partners, with support for various manufacturing applications and agent harnesses.
- What new model support did NVIDIA add?
- NVIDIA says it added day-zero support for Nemotron Bolt E4B 4B, GLM, DeepSeek, Qwen, MiniMax, Liquid AI's LFM2.5-24B-A2B, and the FLUX.3 Klein diffusion model.