---
format: "aidr-story-markdown/v1"
id: "a17ef475e2c86f35818cf1781c926ebaaa824002a91b38342c4325501e76f91c"
canonical_url: "https://aidr.today/a17ef475?lang=en"
title: "When One GPU Isn’t Enough: Serving GLM-5.3 Across Two NVIDIA DGX Stations with vLLM"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-13T17:00:00.000Z"
category: "Frameworks"
topics: ["nvidia","inference","framework"]
source_urls: ["https://developer.nvidia.com/blog/when-one-gpu-isnt-enough-serving-glm-5-3-across-two-nvidia-dgx-stations-with-vllm/"]
summary: "The NVIDIA GLM-5.3 NVFP4 model has 753 billion parameters with 40 billion activated per token. Two NVIDIA DGX Stations connected over ConnectX-8 RDMA serve the model using tensor parallelism with TP=2. Station A returns the answer through one OpenAI-compatible API."
---

# When One GPU Isn’t Enough: Serving GLM\-5\.3 Across Two NVIDIA DGX Stations with vLLM

> [Open the canonical story](<https://aidr.today/a17ef475?lang=en>)

**Published:** 2026-11-13T17:00:00.000Z
**Category:** Frameworks
**Topics:** nvidia, inference, framework

## Summary

The NVIDIA GLM\-5\.3 NVFP4 model has 753 billion parameters with 40 billion activated per token\. Two NVIDIA DGX Stations connected over ConnectX\-8 RDMA serve the model using tensor parallelism with TP\=2\. Station A returns the answer through one OpenAI\-compatible API\.

## Sources

- [Story source](<https://developer.nvidia.com/blog/when-one-gpu-isnt-enough-serving-glm-5-3-across-two-nvidia-dgx-stations-with-vllm/>)