Dell updates PowerScale, ObjectScale to accelerate AI Factory rollout – Blocks and Files

Dell is refreshing its PowerScale and ObjectScale storage systems as part of a slew of AI Factory announcements on the first day of the Dell Technologies World conference.

The company positions its storage systems, data lakehouse, servers, and so on as integrated parts of an AI Factory set of offerings closely aligned with Nvidia’s accelerators and providing a GenAI workflow capability as well as supporting traditional apps. PowerScale – formerly known as Isilon – is its scale-out, clustered filer node offering. ObjectScale is distributed, microservices-based, multi-node, scale-out, and multi-tenant object storage software with a single global namespace that supports the S3 API.  

Jeff Clarke, Dell Technologies COO, stated: “It has been a non-stop year of innovating for enterprises, and we’re not slowing down. We have introduced more than 200 updates to the Dell AI Factory since last year. Our latest AI advancements – from groundbreaking AI PCs to cutting-edge data center solutions – are designed to help organizations of every size to seamlessly adopt AI, drive faster insights, improve efficiency and accelerate their results.” 

As background, Nvidia has announced an extension of its storage server/controller host CPU+DRAM bypass NVMe/RDMA-based GPUDirect file protocol to S3, so that object data can be fed fast to its GPUs using similar, RDMA-based technology. Parallel access file systems like IBM’s Storage Scale, Lustre, VAST Data, VDURA, and WEKA have a speed advantage over serial filers, even scale-out ones, like PowerScale and Qumulo. Dell has responded to this with its Project Lightning initiative.

With these points in mind, the ObjectScale product is getting a denser version along with Nvidia BlueField-3 DPU (Data Processing Unit)  and Spectrum-4 networking support. BlueField 3 is powered by ARM processors and can run containerized software such as ObjectScale. Spectrum-4 is an Ethernet platform product providing 400Gbit/s end-to-end connectivity. Its components include a Spectrum-4 switch, ConnectX-7 SmartNIC, BlueField-3, and DOCA infrastructure software.

The denser ObjectScale system will support multi-petabyte scale andis built from PowerEdge R7725xd server nodes with 2 x AMD EPYC gen 5 CPUs, launching June 2025. It will offer the highest storage density NVMe configurations in the Dell PowerEdge portfolio. The system will feature Nvidia-enhanced BlueField-3 DPUs and Spectrum-4 Ethernet switches and provide planned network connectivity of up to 800 Gb per second.

The vendor says ObjectScale will support S3 over RDMA, making unstructured data stored as objects available much faster for AI training and inferencing. It has up to 230 percent higher throughput and up to 80 percent lower latency. It will also provide 98 percent reduced CPU load compared to traditional S3 data transfers, it claims. A fully managed S3 Tables feature, supporting open table formats that integrate with AI platforms, will be available later this year.

PowerScale gets S3 Object Lock WORM in an upcoming release, along with S3 bucket logging and protocol access logging. PowerScale file-to-object SmartSync automates data replication directly to AWS, Wasabi or Dell ObjectScale for lower-cost back up storage, and can burst to the cloud using EC2 for compute-heavy applications.

A  PowerScale Cybersecurity Suite is an AI, software-driven product designed to provide ransomware detection, minimum downtime when a threat occurs, and near-instant recovery. There are 3 bundles:

• Cybersecurity software for real-time ransomware detection and mitigation, including a full audit trail when an attack occurs. 
• Airgap vault for immutable backups. 
• Disaster recovery software for seamless failover and recovery to guarantee business continuity. 

Project Lightning is claimed to be “the world’s fastest parallel file system per new testing, delivering up to two times greater throughput than competing parallel file systems.” This is according to internal and preliminary Dell testing comparing random and sequential throughput per rack unit. Dell has not provided specific throughput figures, which makes independent comparison difficult. A tweet suggested 97Gbps throughput. The company says Project Lightning will accelerate training time for large-scale and complex AI workflows.

Dell says Lightning is purpose-built for the largest AI deployments with tens of thousands of GPUs. Partners such as WWT and customers such as Cambridge University are active participants in  a multi-phase customer validation program, which includes performance benchmarking, feature testing, and education to drive product requirements and feedback into the product. 

Dell is introducing a high-performance offering built with PowerScale, Project Lightning and PowerEdge XE servers. It will use KV cache and integrate Nvidia’s Inference Xfer Library (NIXL), part of Nvidia’s Dynamo offering, making it ideal for large-scale, complex, distributed inference workloads, according to Dell. Dynamo serves generative AI models in large-scale distributed environments and includes optimizations specific to large language models (LLMs), such as disaggregated serving and key-value cache (KV cache) aware routing.

A Dell slide shows Project Lightning sitting as a software layer above both ObjectScale and PowerScale in an AI Data Platform concept:

We asked about this, and Dell’s Geeta Vaghela, a senior product management director, said: ”We really start to see parallel file systems not being generic parallel file systems, but really optimised for this AI use case and workflow.” She envisages it integrating with KV cache. Dell is now looking to run private previews of the Project Lightning software.

Dell says its AI Data Platform updates improve access to high quality structured, semi-structured, and unstructured data across the AI life cycle. There are Dell Data Lakehouse enhancements to simplify AI workflows and accelerate use cases, such as recommendation engines, semantic search, and customer intent detection by creating and querying AI-ready datasets. Specifically the Dell Data Lakehouse gets:

• Native Vector Search Integration in the Dell Data Analytics Engine, powered by Starburst, bringing semantic understanding directly into SQL workflows and bridging the gap between structured query processing and unstructured data exploration.
• Hybrid Search builds on the vector search capability by combining semantic similarity with traditional keyword matching, within a single SQL query.
• Built-In LLM Functions integrate tools like text summarization and sentiment analysis into SQL-based workflows. 
• Automated Iceberg Table Management looks after maintenance tasks such as compaction and snapshot expiration. 

There are PowerEdge server and network switch updates as part of this overall Dell AI Factory announcement. Dell is announcing Managed Services for the Dell AI Factory with Nvidia to simplify AI operations with management of the full Nvidia AI solutions stack, including AI platforms, infrastructure, and Nvidia AI Enterprise software. Dell managed services experts will handle 24×7 monitoring, reporting, version upgrades, and patching.

The Nvidia AI Enterprise software platform is now available directly from Dell, and customers can use Dell’s AI Factory with Nvidia NIM, NeMo microservices, Blueprints, NeMo Retriever for RAG, and Llama Nemotron reasoning models. They can, Dell says, “seamlessly develop agentic workflows while accelerating time-to-value for AI outcomes.” 

Dell AI Factory with Nvidia offerings support the Nvidia Enterprise AI Factory validated design, featuring Dell and Nvidia compute, networking, storage, and Nvidia AI Enterprise software. This provides an end-to-end, fully integrated AI product for enterprises. Red Hat OpenShift is available on the Dell AI Factory with Nvidia.

Availability

• Dell Project Lightning is available in private preview for select customers and partners now. 
• Dell Data Lakehouse updates will be available beginning in July 2025. 
• Dell ObjectScale with NVIDIA BlueField-3 DPU and Spectrum-4 Ethernet Switches is targeting availability in 2H 2025. 
• The Dell high-performance system built with Dell PowerScale, Dell Project Lightning and PowerEdge XE servers is targeting availability later this year. 
• Dell ObjectScale support for S3 over RDMA will be available in 2H 2025. 
• The NVIDIA AI Enterprise software platform, available directly from Dell, will be available May 2025. 
• Managed Services for Dell AI Factory with NVIDIA are available now.