Bentaus has deployed DriveNets AI Fabric as the back-end network for a new AI cluster running AMD Instinct MI325X GPUs, pairing it with Bentaus’ Ziani orchestration software to reduce energy use per AI token. In the deployment, Bentaus reports a 30% cut in GPU power while maintaining 97% of token throughput.
The cluster combines DriveNets’ Ethernet-based fabric with Ziani Systems platform capabilities branded GPU-FLX and CPU-FLX, which manage compute power in real time. Bentaus describes the goal as running AI compute as a flexible load, curtailing GPU and CPU power consumption while keeping workloads running, with the intent of making the infrastructure more responsive to grid conditions.
DriveNets says its AI Fabric performs “full-stack networking optimization—from switch, via NIC, to RCCL and kernel optimization,” and that it is optimized for AMD environments based on a reference architecture jointly engineered by DriveNets and AMD. DriveNets also points to published benchmarks for the architecture, citing approximately 5% higher throughput and 12% better time to first token than publicly available results, while minimizing power consumption and lowering effective energy costs.
For data center operators, the practical angle is straightforward: if the power layer and the network layer can be tuned together without stalling GPUs, you can target more usable output per watt without treating the cluster as a fixed, non-negotiable load. That’s directly relevant in environments where power availability, not floor space, is the limiting factor for bringing AI capacity online.
“Ziani flexes our clusters acknowledging grid conditions in real time, and DriveNets’ fabric makes sure the GPUs we run are never waiting on the network,” said Bob Davidoff, CEO and Founder of Bentaus. DriveNets CEO and Founder Ido Susan said, “The next race is to produce [tokens] on less power.”
Source: DriveNets
















