{"id":3754,"date":"2026-03-31T18:12:18","date_gmt":"2026-03-31T18:12:18","guid":{"rendered":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/2026\/03\/31\/run-and-iterate-on-llms-faster-with-docker-model-runner-on-dgx-station\/"},"modified":"2026-03-31T18:12:18","modified_gmt":"2026-03-31T18:12:18","slug":"run-and-iterate-on-llms-faster-with-docker-model-runner-on-dgx-station","status":"publish","type":"post","link":"https:\/\/rssfeedtelegrambot.bnaya.co.il\/index.php\/2026\/03\/31\/run-and-iterate-on-llms-faster-with-docker-model-runner-on-dgx-station\/","title":{"rendered":"Run and Iterate on LLMs Faster with Docker Model Runner on DGX Station"},"content":{"rendered":"

Back in October, we showed how Docker Model Runner on the NVIDIA DGX Spark<\/a> makes it remarkably easy to run large AI models locally with the same familiar Docker experience developers already trust. That post struck a chord: hundreds of developers discovered that a compact desktop system paired with Docker Model Runner could replace complex GPU setups and cloud API calls.<\/p>\n

Recently at NVIDIA GTC 2026, NVIDIA is raising the bar with NVIDIA DGX Station <\/a>and we\u2019re excited to add support for it in Docker Model Runner!\u00a0 The new DGX Station brings serious performance, and Model Runner helps make it practical to use day to day. With Model Runner, you can run and iterate on larger models on a DGX Station, using the same intuitive Docker experience you already know and trust.<\/p>\n

From NVIDIA DGX Spark to DGX Station: What has changed and why does this matter?<\/h2>\n

NVIDIA DGX Spark<\/a>, powered by the GB10 Grace Blackwell Superchip, gave developers 128GB of unified memory and petaflop-class AI performance in a compact form factor. A fantastic entry point for running models.<\/p>\n

NVIDIA DGX Station is a different beast entirely. Built around the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip<\/a>, it connects a 72-core NVIDIA Grace CPU and NVIDIA Blackwell Ultra GPU through NVIDIA NVLink-C2C, creating a unified, high-bandwidth architecture built for frontier AI workloads. It brings data-center-class performance to a deskside form factor. Here are the headline specs:<\/p>\n

\n\n\n\n\n\n\n\n\n
\n<\/th>\n\n

DGX Spark (GB10)<\/p>\n<\/th>\n

\n

DGX Station (GB300)<\/p>\n<\/th>\n<\/tr>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

GPU Memory<\/p>\n


\n \n <\/p><\/td>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

128 GB unified<\/p>\n


\n \n <\/p><\/td>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

252 GB<\/p>\n


\n \n <\/p><\/td>\n<\/tr>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

GPU Memory Bandwidth<\/p>\n


\n \n <\/p><\/td>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

273 GB\/s<\/p>\n


\n \n <\/p><\/td>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

7.1 TB\/s<\/p>\n


\n \n <\/p><\/td>\n<\/tr>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

Total Coherent Memory<\/p>\n


\n \n <\/p><\/td>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

128 GB<\/p>\n


\n \n <\/p><\/td>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

748 GB<\/p>\n


\n \n <\/p><\/td>\n<\/tr>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

Networking<\/p>\n


\n \n <\/p><\/td>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

200 Gb\/s<\/p>\n


\n \n <\/p><\/td>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

800 Gb\/s<\/p>\n


\n \n <\/p><\/td>\n<\/tr>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

GPU Architecture<\/p>\n


\n \n <\/p><\/td>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

Blackwell (5th-gen Tensor Cores, FP4)<\/p>\n


\n \n <\/p><\/td>\n

\n <\/span>\n


\n <\/span><\/span><\/p>\n

Blackwell Ultra (5th-gen Tensor Cores, FP4)<\/p>\n


\n \n <\/p><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

With 252GB of GPU memory at 7.1 TB\/s of bandwidth and a total of 748GB of coherent memory, the DGX Station doesn\u2019t just let you run frontier models,\u00a0 it lets you run trillion-parameter models, fine-tune massive architectures, and serve multiple models simultaneously, all from your desk.<\/p>\n

Here\u2019s what 748GB of coherent memory and 7.1 TB\/s of bandwidth unlock in practice:<\/p>\n