Dell AI Factory

from Jeff Fonke

Dell is at the forefront of innovation with its AI factory approach. This comprehensive strategy encompasses high-performance architecture (HPA) accelerators, networking and storage, along with the software, MLOps platforms and integrations with top software suites like NVIDIA AI Enterprise (NVAIE) and OpenShift AI platform. 

One exciting development is the upcoming release of the Dell XE9680L, featuring Blackwell B200 GPUs. This server is a crucial component of the reference architecture. Notably, the 'L' in its name signifies liquid cooling. Its predecessors have already been deployed in our AI Proving Ground and are ready for testing. 

Additionally, we are working toward the build-out of the data analytics engine (DDAE 660) and integrating that with the best-in-class SuperPOD-certified storage from Dell's PowerScale lineup. As customers explore data pipelines and delve into HPA-related use cases like generative AI, computer vision, machine learning and other applied AI use cases, WWT can showcase the value of Dell's AI factory in our AI Proving Ground. 

from Matt Halcomb

The announcement of the XE9680L (the newly designed system that supports 8 x Blackwell GPUs in 4 rack unit space with direct liquid cooling capabilities) was great to hear. And, the news that it will support NVIDIA Blackwell GPUs along with solutions from AMD and Intel shows the value and diversity that Dell brings to the AI game. 

Dell also announced turnkey rack solutions, available in three different configuration sizes: a 70KW air-cooled solution, a 100KW liquid-cooled solution with 72 Blackwell GPUs, and a 130KW liquid-cooled solution focused on the NVIDIA BH200, along with solutions from Intel and AMD. 

In addition, the Dell AI Factory alignment with NVIDIA includes integrating Dell's NativeEdge Zero Trust Edge solution with NVIDIA NIMs (NVIDIA Inference Microservices). This integration empowers customers to create NativeEdge Blueprints, which can be securely deployed to any Dell NativeEdge appliance at the Edge.

Lastly, Dell introduced the Dell Enterprise Hub on Hugging Face, a powerful solution enabling organizations to train and deploy open, customized large language models (LLMs) on-premises using Dell infrastructure. This streamlined experience accelerates time-to-value for customers building AI applications, such as chatbots and customer support, directly within Hugging Face's leading open platform for AI builders.

Dell Servers

from Steven McCall

Dell recently announced the PowerEdge XE9680L, which features 8 NVIDIA Blackwell GPUs. This server comes in a compact 4U form factor and incorporates liquid cooling for efficient performance. Additionally, it boasts 400GB networking from Broadcom to enhance data throughput.

Regarding AI, there remains a large interest in inference and RAG (retrieval-augmented generation). Organizations can leverage pre-trained models supplemented with their private data for various AI workloads. These areas will continue to grow with the Dell Enterprise Hub and focus on Dell AI factories.

Furthermore, Dell's existing partnership with Nutanix expands significantly with the introduction of Nutanix Cloud Platform on Dell PowerEdge and Nutanix Cloud Platform with Dell PowerFlex. While Nutanix traditionally relies on its own software-defined storage, the new offering with PowerFlex storage provides greater flexibility and scalability options for the platform. 

Dell Storage

from John Lochausen

While Michael Dell and Jeff Clarke touched on this briefly in their keynotes, Dell released some important updates to its storage lineup. 

First, the launch of PowerScale F910 completes this year's update to the PowerScale all-NVMe lineup, following the earlier F710 and F210. This takes the flagship product into Dell's 16G server platform, based on PowerEdge R760. Notable improvements in performance are powered by the significant hardware updates and code improvements available in OneFS 9.8.  Designed for high-end workloads, the F910 is particularly AI-friendly, ensuring GPUs remain well-fed with data. Dell touted PowerScale as the first Ethernet-based storage solution to be SuperPod certified. WWT was a beta site for F910.

Additionally, PowerStore Prime took center stage in several breakout sessions. This exciting development combines the release of PowerStoreOS 4.0 with program updates, including a data reduction guarantee that has been improved from 4:1 to 5:1.

Here are the PowerStore Prime key highlights:

  • The OS update significantly improves metro replication support, now extending to both Windows and Linux environments and the use of volume groups with metro.
  • Through a partnership with ProLion, PowerStore Prime introduces metrosync capabilities specifically designed for file workloads.
  • Dell launched the 3200Q, a QLC (quad-level cell) model. It shares the same system architecture as the existing 3200T but features QLC drives. This design choice increases storage density and lowers the cost-per-gigabyte while maintaining excellent all-NVMe performance.
  • Users can now seamlessly perform online system upgrades from gen-2 hardware to higher-level gen-2 hardware without disrupting their workloads.

WWT was a beta site for PowerStore 4.0.


We only touched on a sampling of everything announced at DTW 2024, but it's already shaping up to be an exciting year for WWT and Dell! Thanks for reading.