Read the full transcript
Hi, I’m Arvind Balakumar. I’m the Corporate Vice President of Global Cluster Engineering at AMD.
Compute in an AI factory is only as valuable as the network’s ability to deliver it, and that’s where DriveNets really stood out for AMD. So together with DriveNets, we validated an open Ethernet-based reference architecture that pairs AMD’s Instinct 350 series GPUs along with DriveNets’ AI fabric. And through this, you’re able to maximize GPU utilization, you improve workload efficiency, and you dramatically simplify deployment across both training and inference, which is a big pain point for our customers.
And this gives our customers real optionality, not just a single supplier lock-in. In addition to this, our rigorous resiliency testing of this reference architecture showed consistent collective communication performance under concurrent RDMA traffic with no degradation during transient link disruptions. So this reference architecture isn’t just faster in a benchmark, it is highly dependable under real production workloads. The DriveNets AI Fabric delivered roughly 5% higher throughput, a 12 to 16% reduction in time to first token, and sub-20 millisecond inter-token latency. All at multi-node scale.
This is the future that AMD and DriveNets are building together. A production-ready, highly performant, and an open architecture that adds significant value to our customers in the AI infrastructure app.
Want to learn how DriveNets is reshaping AI infrastructure?
Explore DriveNets AI Fabric full stack networking solutions for AI Infrastructure
