Dell expanded its AI strategy with new partnerships, as well as upgraded hardware and software, giving customers another choice for building and deploying their AI workloads.
This month, Dell will add more performance and density to its unstructured data storage system, PowerScale, as well as update its underlying software, PowerScale OneFS. PowerScale is set to become the first Ethernet-connected storage device to be validated on Nvidia DGX SuperPod, Nvidia's data center reference architecture for high-performance computing and AI workloads.
Dell is also expanding its AI partnership with AMD; its PowerEdge XE9680 server will support the AMD Instinct MI300X accelerator, a CPU-GPU design for AI workloads that became available Wednesday. News of the expanded partnership came on the heels of Dell's third-quarter 2024 earnings call, where Jeff Clarke, Dell COO, acknowledged a 39-week lead time for Nvidia-based systems.
One week earlier, Dell shared that it was partnering with AI lab Imbue to develop new AI training models on PowerEdge XE9680 servers with Nvidia H100 Tensor Core GPUs. The Imbue partnership adds to Dell's mid-November news of building a portal to AI platform provider Hugging Face.
Dell's focus on AI -- from storage to training models -- suggests a push to provide competitive technology to the major cloud providers, where generative AI has taken root, according to Steve McDowell, an analyst and founding partner at NAND Research.
Steve McDowellAnalyst and founding partner, NAND Research
"What Dell is trying to do is put the pieces together to say, 'We can offer everything the cloud guys can deliver in terms of full stack,'" he said.
But Dell is offering this stack so that customers can run AI on premises, McDowell said.
PowerScale to power AI
In the first quarter of 2024, PowerScale will be upgraded with new all-flash storage based on Dell's PowerEdge servers, doubling its performance. Its OS, OneFS, will also be updated, according to the vendor. Both hardware and software upgrades aim to increase the ease of AI modeling and inferencing.
Increasing storage's streaming performance, both in reads and in writes, is important in AI as users don't want data to stall when it comes off storage, McDowell said.
"What they've done is essentially optimized [OneFS] for AI and the access patterns of AI," he said.
The focus on optimizing PowerScale for AI is a logical step for Dell, as file-based data is a popular storage type for AI workloads including videos and documents, according to Dave Raffo, an analyst at Futurum Group.
Plus, Dell has a strong foothold in the file storage market as PowerScale was originally Isilon, which launched more than 20 years ago and has been growing since then, Raffo said. It also keeps Dell competitive with another on-premises player: HPE.
Last week at HPE Discover Barcelona, the vendor unveiled a doubling of capacity of its file storage service in its GreenLake offering. It also added faster networking with Nvidia Quantum-2. Both Dell and HPE have their eyes on AI with these updates, with Dell having potentially more file storage customers but older technology, while HPE uses the younger Vast Data for file storage, Raffo said.
With unstructured data and generative AI, the storage depends on the use case, according to Scott Sinclair, an analyst at TechTarget's Enterprise Strategy Group. He added that customers can address smaller projects on premises with products such as Dell's PowerScale.
"Dell is well equipped, not just with PowerScale, but with its breadth of storage options it offers, because people may not need petabytes' worth of data for a project," Sinclair said.
Ethernet and SuperPod
PowerScale is on track to become the first Ethernet-connected storage system to be validated with Nvidia DGX SuperPod, according to Dell. Validation is expected to be completed in the first quarter of 2024. This will give customers access to the Nvidia AI Enterprise software platform, Dell said. Previous to this, SuperPod validation was for InfiniBand-connected equipment.
The expansion of DGX SuperPod to PowerScale is good for Dell customers due to the ubiquity of Ethernet, Sinclair said. "If you're trying to make AI more accessible, InfiniBand may not be the most accessible option."
McDowell echoed the point, saying Dell customers won't have to install Fibre Channel or InfiniBand to work with high-performance workloads on premises.
"Leveraging Ethernet allows almost a plug-and-play into existing infrastructure," he said, adding that the industry is moving in this direction and Dell is a little ahead of the game.
AMD as an alternative
In an expansion of the Dell Generative AI portfolio, the vendor is partnering with AMD to bring the Instinct MI300X accelerator GPU to the Dell PowerEdge XE9680 server. Dell is offering both AMD and Nvidia GPU options to customers, with the AMD version expected in the first half of 2024.
McDowell said the long lead time to acquire Nvidia GPUs is partly why Dell expanded to AMD GPUs. He added that the MI300 series also performs well.
Historically, Nvidia Compute Unified Device Architecture locked out AI competitors, but that changed with generative AI and large language models, McDowell said. While there wasn't much choice in traditional AI, there is choice with generative AI, and Dell is working to offer its customers that choice.
Doing so also means providing customers with more nuance, as not everyone needs the performance capabilities that Nvidia GPUs offer, Sinclair said. Nvidia offers top-level performance, which is in high demand for hyperscalers and where demand outstrips supply. Working with AMD could help Dell address demand for smaller, on-premises use cases, which also exists.
"Partnering with AMD makes sure that people can get supplied when they need it to start their projects," Sinclair said.
Adam Armstrong is a TechTarget Editorial news writer covering file and block storage hardware and private clouds. He previously worked at StorageReview.com.