deepseek-v4.1IMAGE
DeepSeek V4.1 Flash on 4 DGX Sparks: second RoCE cable lifts long-prompt reads 15%
Too large for one DGX Spark, the model is split across four nodes swapping work over RoCE each step; a second cable doubles the lanes and speeds long-prompt reads by about 15%.
Tech2Wild
@Tech2WildArchived 12 days agoOriginally posted 2026.10.18

Tech2Wild @Tech2WildOriginally posted 2026.10.18
🧠 Why a 2nd cable = more speed: DeepSeek V4.1 Flash is too big for one DGX Spark, so it's split across 4. Every step, the Sparks swap their work over RoCE. 1 cable = 1 lane. 2 cables = double the lanes 🛣️ Long prompts move the most data, so reading them jumped +15% 📖
This post quotes another post: “DeepSeek V4.1 Flash on dual-cable DGX Spark: 93.6 tok/s single user”
x.com/tech2wild/status/2111666660101026191Archived 2026.10.18 16:25