ServerMO5d ago

Optimize AI Cluster Networks with Multi-Rail RoCEv2

Standard Ethernet stalls GPU training with packet drops and ECMP hash collisions. Master the SRE fabric playbook: Bypass the OS kernel with RDMA, enforce lossless PFC (use watchdogs to prevent deadlocks!), and use Multi-Rail PCIe affinity to dedicate physical NICs directly to GPUs.

Read the bare-metal architecture guide by @ServerMO:
🔗 https://www.servermo.com/blogs/multi-rail-rocev2-ai-cluster/

#SRE #DevOps #AI #Networking #BareMetal #RoCEv2 #MachineLearning

ServeTheHomeMay 6
NVIDIA is talking about is Spectrum-X MRC, a custom RDMA transport protocol already powering frontier gigascale AI deployments#NVIDIA #RDMA #RoCEv2 #Spectrum-X
NVIDIA Spectrum-X Ethernet MRC is the Custom RDMA Transport Protocol for Gigascale AI
NVIDIA Spectrum-X Ethernet MRC is the Custom RDMA Transport Protocol for Gigascale AI

NVIDIA is talking about is Spectrum-X MRC, a custom RDMA transport protocol already powering frontier gigascale AI deployments

ServeTheHome
ServeTheHomeMay 6
NVIDIA is talking about is Spectrum-X MRC, a custom RDMA transport protocol already powering frontier gigascale AI deployments#NVIDIA #RDMA #RoCEv2 #Spectrum-X
NVIDIA Spectrum-X MRC is the Custom RDMA Transport Protocol for Gigascale AI
NVIDIA Spectrum-X MRC is the Custom RDMA Transport Protocol for Gigascale AI

NVIDIA is talking about is Spectrum-X MRC, a custom RDMA transport protocol already powering frontier gigascale AI deployments

ServeTheHome
Peter J. WelcherFeb 27, 2025
New blog “How AI Ate My Blog on RoCEv2”. #PeterWelcher #CCIE1773 #AI #ECN #PFC #RoCEV2 #Infiniband. URL: https://www.linkedin.com/pulse/ai-ate-my-blog-rocev2-peter-welcher-uu9ue/.
AI Ate My Blog on RoCEv2

I acknowledge I’ve been a blog technology summarizer for quite a while. It served to help me broaden/solidify my skills and hopefully help others do so as well.

AsterfusionJun 12, 2024
Asterfusion CX-N switches utilize ROCE technology to deliver top-notch performance comparable to expensive InfiniBand switches, all at a fraction of the cost. This article provides a comprehensive overview of HPC fundamentals and the workings of ROCE technology, alongside a thorough comparison of test results between Asterfusion ROCE data switches and IBs in an HPC scenario. The final segment includes a visual guide on how to effectively configure #ROCEv2 on Asterfusion SONiC data centre switches.
https://cloudswit.ch/blogs/roce-for-hpc-test-data-and-deploy-on-sonic/
RoCE Technology For HPC- Test Data & Practical Implementations On SONiC Switch - Asterfusion Data Technologies

This article explores what ROCE is and what high performance computing is, as well as comparing the test data of the Asterfusion SONiC -based ROCE switch and the IB switch in an HPC scenario

Asterfusion Data Technologies
Gestalt ITJul 10, 2023
Just Posted: The article explores Cisco's Data Center Networking Blueprint for AI/ML applications, emphasizing the need for low-latency, lossless networks and discussing the two types of AI clusters and their network requirements. It highlights the challenges of scalability and the importance of building robust networks to handle the growing amount of data in AI modeling.
https://gestaltit.com/tech-field-day/sulagna/designing-a-lossless-ai-ml-network-with-cisco-data-center-networking-blueprint/
#AI #CiscoLiveUS #DataCenterNetworking #ML #RoCEv2 #TFDx
Designing a Lossless AI/ML Network with Cisco Data Center Networking Blueprint - Gestalt IT

In this Tech Field Day Extra article from Cisco Live Sulagna Saha discusses the Cisco Data Center Networking Blueprint for AI/ML applications.

Gestalt IT