<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Rdma on 5glinux</title><link>https://blog.5glinux.com/tags/rdma/</link><description>Recent content in Rdma on 5glinux</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Mon, 19 Oct 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://blog.5glinux.com/tags/rdma/index.xml" rel="self" type="application/rss+xml"/><item><title>Running DeepSeek V4 Flash on two DGX Sparks with Kubernetes and Codex CLI</title><link>https://blog.5glinux.com/posts/deepseek-v4-dgx-spark-codex/</link><pubDate>Mon, 19 Oct 2026 00:00:00 +0000</pubDate><guid>https://blog.5glinux.com/posts/deepseek-v4-dgx-spark-codex/</guid><description>DeepSeek V4 Flash needs more memory than one DGX Spark has. This guide splits it across two Sparks with vLLM tensor parallelism over RoCE, reaches 51 tokens per second, and connects OpenAI&amp;rsquo;s Codex CLI as a fully local coding agent.</description></item><item><title>Longhorn v2 on SPDK: TCP catches up, RDMA still wins on latency (part 1)</title><link>https://blog.5glinux.com/posts/longhorn-v2-spdk-rdma-vs-tcp/</link><pubDate>Mon, 12 Oct 2026 00:00:00 +0000</pubDate><guid>https://blog.5glinux.com/posts/longhorn-v2-spdk-rdma-vs-tcp/</guid><description>Two snapshots of Longhorn&amp;rsquo;s SPDK-based v2 engine on the same three Ampere nodes: 1.12 with my RDMA patches, then 1.14 development builds with a newer SPDK. NVMe/TCP got about 7× lower write latency and now fills a 25 GbE link. RDMA still halves write latency.</description></item></channel></rss>