---
format: "aidr-story-markdown/v1"
id: "c170fbdba4e585dcdc5abd493a1b0fbe9dee2416138613ef137843622522b446"
canonical_url: "https://aidr.today/c170fbdb?lang=en"
title: "NVIDIA Shows 3 Ways to Speed Up MoE Inference on DGX Spark, Lifting Ling-3.0-flash Throughput 42.5%"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-03T20:28:22.000Z"
category: "Infra"
topics: ["nvidia","inference"]
source_urls: ["https://www.marktechpost.com/2026/11/03/nvidia-dgx-spark-moe-inference-ling-3-0-flash/","https://news.ycombinator.com/item?id=50362905","https://www.mindstudio.ai/blog/linux-nvidia-rtx-spark-laptop"]
summary: "NVIDIA demonstrated three stacked optimizations that speed up MoE inference on a single DGX Spark, lifting GSM8K decode throughput on Ant Group inclusionAI's 124B Ling-3.0-flash by 42.5% with no detectable accuracy change."
---

# NVIDIA Shows 3 Ways to Speed Up MoE Inference on DGX Spark, Lifting Ling\-3\.0\-flash Throughput 42\.5%

> [Open the canonical story](<https://aidr.today/c170fbdb?lang=en>)

**Published:** 2026-11-03T20:28:22.000Z
**Category:** Infra
**Topics:** nvidia, inference

## Summary

NVIDIA demonstrated three stacked optimizations that speed up MoE inference on a single DGX Spark, lifting GSM8K decode throughput on Ant Group inclusionAI's 124B Ling\-3\.0\-flash by 42\.5% with no detectable accuracy change\.

## Sources

- [Story source](<https://www.marktechpost.com/2026/11/03/nvidia-dgx-spark-moe-inference-ling-3-0-flash/>)
- [Discussion](<https://news.ycombinator.com/item?id=50362905>)
- [Story source](<https://www.mindstudio.ai/blog/linux-nvidia-rtx-spark-laptop>)