Home > AI Solutions > Artificial Intelligence > White Papers > White Paper - Virtualizing GPUs for AI with VMware and NVIDIA Based on Dell Infrastructure > Dell Technologies
The latest Dell PowerEdge servers are certified for VMware vSphere 8 and vSAN 8. PowerEdge Intel-based servers use the latest 4th Generation Intel Xeon Scalable processors. Other key features include:
PowerEdge AMD-based servers use the latest EPYC 4th generation processors 4th Gen AMD EPYC processors. Other key features include:
The following table lists the PowerEdge servers that are supported with NVIDIA AI Enterprise and the number of Ampere GPUs that are supported with each server model:
Table 2. Supported PowerEdge servers
Maximum A100 GPUs and A30 GPUs | |
PowerEdge R760 | 2 |
PowerEdge R7625 | 2 |
These PowerEdge servers are NVIDIA-Certified Systems and have been proven through a rigorous suite of functional and performance tests. The test results confirm that these servers can deliver high performance both in single-node and networked multinode cluster training and inference benchmarks. Also, these servers are certified to be compatible with NVIDIA AI Enterprise through additional testing and validation.
PowerScale storage helps unlock the structure within data and address the challenges of unstructured data management. PowerScale is the next evolution of OneFS—the operating system powering the scale-out NAS platform. The PowerScale family includes Dell Isilon nodes and PowerScale nodes, with PowerScale OneFS running across all of them. The software-defined architecture of OneFS provides simplicity at scale, intelligent insights, and the ability to have data anywhere it needs to be. Whether hosting file shares or home directories, or delivering high-performance data access for applications such as analytics, video rendering, and life sciences, PowerScale can seamlessly scale performance, capacity, and efficiency to handle any unstructured data workload. The new PowerScale all-flash platforms co-exist seamlessly in the same cluster with your existing Isilon nodes to drive your traditional and modern applications.
In this validated design, we use PowerScale as storage for the data lake—the data repository for unstructured data that you can use for neural network training. PowerScale All-Flash Scale-out NAS storage is ideal, delivering the analytics performance and extreme concurrency at scale to consistently feed the most data-hungry deep learning algorithms.
GPUDirect Storage, an NVIDIA technology, enables a direct data path between local or remote storage, like NVMe or NVMe over Fabric (NVMe-oF), and GPU memory. GPUDirect Storage avoids extra copies through a bounce buffer in the CPU’s memory. It enables a direct memory access (DMA) engine near the storage to move data on a direct path into or out of GPU memory – all without burdening the CPU or GPU. PowerScale supports GPUDirect Storage.
PowerStore is a modern storage appliance designed for the data era. The single architecture of PowerStore for block, file, and VMware vVols uses the latest technologies to support an enterprise-class variety of traditional and modern workloads – from relational databases, to ERP and EMR applications, cloud native applications, and file-based workloads such as content repositories and home directories. The ability to accommodate application, multiprotocol network, and multiformat storage diversity (physical and virtual volumes, containers, and traditional files) in a single 2U appliance provides business-enabling flexibility and helps IT simplify and consolidate their infrastructure.
Administrators can choose to deploy PowerStoreOS in a bare-metal configuration directly on the PowerStore hardware or in a VM running on PowerStore’s optional integrated VMware hypervisor, providing yet another layer of isolation, intelligence, and abstraction.
In this validated design, we use the following Dell switches: