NVIDIA DGX Spark 和 DGX Station 将最新的开源与前沿模型带到桌面端

January 5, 2026byChris Marriott2026年1⽉5⽇，Chris Marriott Open-source AI is accelerating innovation across industries, andNVIDIA DGX SparkandDGX Stationarebuilt to help developers turn innovation into impact. 开源AI正在加速各⾏各业的创新，⽽NVIDIA DGX Spark和DGX Station的打造正是为了帮助开发者将创新转化为实际影响。 NVIDIA today unveiled at the CES trade show how the DGX Spark and DGX Station deskside AIsupercomputers let developers harness the latest open and frontier AI models on a local deskside system,from 100-billion-parameter models on DGX Spark to 1-trillion-parameter models on DGX Station. NVIDIA今⽇在CES消费电⼦展上展示了DGX Spark和DGX Station桌边式AI超级计算机，说明它们如何让开发者在本地桌边系统上利⽤最新的开源和前沿AI模型——从DGX Spark上的千亿参数模型，到DGX Station上的万亿参数模型。 Powered by the NVIDIA Grace Blackwell architecture, with large unified memory and petaflop-level AIperformance, these systems give developers new capabilities to develop locally and easily scale to thecloud. 这些系统基于NVIDIA Grace Blackwell架构构建，配备⼤容量统⼀内存和千万亿次级AI性能，使开发者能够获得全新的本地开发能⼒，并轻松扩展⾄云端。 Advancing Performance Across Open-Source AI Models 推动开源AI模型性能提升 A breadth of highly optimized open models that would’ve previously required a data center to run can nowbe accelerated at the desktop on DGX Spark and DGX Station, thanks to continual advancements in modeloptimization and collaborations with the open-source community. 由于模型优化的持续进步以及与开源社区的合作，过去需要在数据中⼼运⾏的⼤量⾼度优化的开放模型，如今可通过DGX Spark和DGX Station在桌⾯端实现加速运⾏。 Preconfigured with NVIDIA AI software andNVIDIA CUDA-Xlibraries, DGX Spark provides powerful, plug-and-play optimization for developers, researchers and data scientists to build, fine-tune and run AI. DGX Spark预配置了NVIDIA AI软件和NVIDIA CUDA-X库，可为开发者、研究⼈员和数据科学家提供强⼤、即插即⽤的优化能⼒，⽤于构建、微调和运⾏AI。 Spark provides a foundation for all developers to run the latest AI models at their desk; Station enablesenterprises and research labs to run more advanced, large-scale frontier AI models. The systems supportrunning the latest frameworks and open-source models — including the recently announcedNVIDIANemotron 3 models— right from desktops. Spark为所有开发者在办公桌前运⾏最新AI模型奠定了基础；Station则使企业和研究实验室能够运⾏更先进、更⼤规模的前沿AI模型。这些系统⽀持直接在桌⾯端运⾏最新的框架和开源模型——其中包括недавно发布的NVIDIA Nemotron 3模型。 The NVIDIA Blackwell architecture powering DGX Spark includes the NVFP4 data format, which enables AImodels to be compressed by up to 70% and boosts performance without losing intelligence. 为DGX Spark提供动⼒的NVIDIA Blackwell架构包含NVFP4数据格式，可将AI模型压缩⾼达70%，并在不损失智能的情况下提升性能。 NVIDIA’s collaborations with the open-source software ecosystem, such as its work with llama.cpp, ispushing performance further, delivering a 35% performance uplift on average when running state-of-the-artAI models on DGX Spark. Llama.cpp also includes a quality-of-life upgrade that speeds up LLM loadingtimes. NVIDIA与开源软件⽣态系统的合作（例如与llama.cpp的协作）正在进⼀步推动性能提升，使在DGXSpark上运⾏最先进的AI模型时，平均性能提升达到35%。Llama.cpp还包括⼀项提升使⽤体验的改进，可加快LLM的加载时间。 DGX Station, with the GB300 Grace Blackwell Ultra superchip and 775GB of coherent memory with FP4precision, can run models up to 1 trillion parameters — giving frontier AI labs cutting-edge computecapability for large-scale models from the desktop. This includes a variety of advanced AI models includingKimi-K2 Thinking, DeepSeek-V3.2, Mistral Large 3, Meta Llama 4 Maverick, Qwen3 and OpenAI gpt-oss-120b. DGX Station搭载GB300 Grace Blackwell Ultra超级芯⽚，并配备775GB具备FP4精度的⼀致性内存，能够运⾏⾼达1万亿参数的模型——为前沿AI实验室从桌⾯端运⾏⼤规模模型提供尖端的计算能⼒。这其中包括多种先进的AI模型，如Kimi-K2 Thinking、DeepSeek-V3.2、Mistral Large 3、Meta Llama 4Maverick、Qwen3以及OpenAI gpt-oss-120b。 “NVIDIA GB300 is typically deployed as a rack-scale system,” said Kaichao You, core maintainer of vLLM.“This makes it difficult for projects like vLLM to test and develop directly on the powerful GB300 superchip.DGX Station changes this dynamic. By delivering GB300 in a compact, single-system form factor deskside,DGX Station enables vLLM to test and develop GB300-specific features at a significantly lower cost. Thisaccelerates development cycles and makes it easy for vLLM to continuously validate and optimize againstGB300.” “NVIDIA GB300通常以机架级系统的形式部署，”vLLM核⼼维护者游凯超表示。“这使得像vLLM这样的项⽬难以直接在强⼤的GB300超级芯⽚上进⾏测试和开发。DGX Station改变了这⼀局⾯。通过以紧凑的单机桌边形态交付GB300，DGX Station使vLLM能够以显著更低的成本测试和开发GB300专属特性。这加快了开发周期，也让vLLM能够持续针对GB300进⾏验证和优化。” “DGX Station brings data-center-class GPU capability directly into my room,” said Jerry Zhou, communitycontributor to SGLang. “It is powerful enough to serve very large models like Qwen3-235B, test trainingframeworks with large model configurations and develop CUDA kernels with extremely large matrix sizes, alllocally without relying on cloud racks. This dramatically shortens the iteration loop for systems and framework development.” “SGLang社区贡献者Jerry Zhou表示：“DGX Station将数据中⼼级GPU能⼒直接带进了我的房间。它的性能强⼤到⾜以在本地运⾏和服务诸如Qwen3-235B这样的⼤型模型，测试⼤模型配置下的训练框架，以及开发具有超⼤矩阵规模的CUDA内核，完全⽆需依赖云端机架。这极⼤地缩短了系统和框架开发的迭代周期。” NVIDIA will be showcasing the capabilities of DGX Station live at CES, demonstrating: NVIDIA将在CES上现场展示DGX Station的能⼒，演示内容包括： LLM pretraining that moves at a blist