Home > Workload Solutions > Data Analytics > White Papers > Scale AI Training and Fine-Tuning with Dell PowerScale and PowerEdge Servers > NFSoRDMA
Dell PowerScale was among the first storage platforms to support NFSoRDMA and NVIDIA’s GPU Direct Storage (GDS). The low-latency direct transfer of data between storage and GPU memory that GDS enables ensures that Dell PowerScale can keep GPUs utilized while running large, demanding AI models.
PowerScale OneFS supports Remote Direct Memory Access (RDMA) over converged Ethernet (RoCEv2). When paired with NFS to enable NFSoRDMA, data is sent directly from the PowerScale network adapter to the compute or application memory, bypassing CPU and operating systems. Combining NFSoRDMA with NVIDIA’s GPU Direct Storage and MagnumIO software modules, data can be transferred directly between NVIDIA GPU memory and PowerScale storage over a low latency, highly efficient network.