Deepak Vij·Oct 16Inside an NVIDIA AI Cluster: From One B200 to a SuperPODIn my previous role as an applied researcher, I was handed “a machine with 8 B200 GPUs.” I used it for months (to perform…
InMajor DigestbyValentin Podkamennyi·Apr 9UALink 2.0 Advances AI Interconnect TechnologyThe UALink Consortium has released version 2.0 of its specification, aiming to provide an open-source alternative for AI interconnects.
Lalit Somavarapha·Jan 27Optimizing GPU Workload Placement in Kubernetes with NVLink-Aware SchedulingThe hidden performance tax
Samarth Narang·Nov 16, 2025How GPUs Talk: A Practical Guide to Multi-GPU Training and CommunicationAn in-depth exploration of multi-GPU training techniques and the communication patterns that enable efficient distributed machine learning.
NADDOD·Jun 19, 2025Understanding Scale-Up vs. Scale-Out in AI InfrastructureIn the ICT(Information and Communication Technology) domain, the concept of “Scale” is frequently encountered. It typically refers to…
InGPU Kernel Hacking for EngineersbyNikheel Vishwas Savant·May 26, 2025Part 2: Inter-GPU Communication with PCIe, NVLink, and BARsIn Part 1, we explored how threads inside a single GPU communicate using warps, shared memory, and barriers. But what happens when your…
Nikheel Vishwas Savant·May 25, 2025The GPU Communication Stack: What Happens Between GPUs When You Scale AIAs deep learning scales beyond single machines, GPUs are no longer isolated accelerators. They’re part of distributed systems —…
ztex, Tony, Liu·Jan 8, 2025Note about “THE NVLINK-NETWORK SWITCH: NVIDIA’S SWITCH CHIP FOR HIGH COMMUNICATION-BANDWIDTH…This post aims to pick some take-away from the presentation about design of NVLINK
Suvasis M·Jun 24, 2024NVLinkToday’s server GPUs are typically connected by the PCI Express (PCIe) bus, which provides a communication bandwidth of 12 GB per second…
