SC19: HPC and AI Innovations From Dell

At SC19, Dell Technologies, Inc. has introduced several solutions, reference architectures and portfolio advancements all designed to simplify and accelerate customers’ HPC and AI efforts.

Continued adoption of AI to solve real-world problems has spurred growth across the HPC industry. According to a recent report from Hyperion Research, the global HPC industry is expected to grow by 7.1% to more than $39.2 billion by 2023 ⁽¹⁾ and HPC-server based AI is expected to rise by more than 29% from 2018 to 2023, reaching $2.7 billion in 2023. ⁽²⁾

“There’s a lot of value in the data that organizations collect, and HPC and AI are helping organizations get the most out of this data,” said Thierry Pellegrino, VP, HPC, Dell. “We’re committed to building solutions that simplify the use and deployment of these technologies for organizations of all sizes and at all stages of deployment.“

Advances storage solutions for HPC
The company is expanding its portfolio of Dell EMC Ready Solutions for HPC Storage with turnkey solutions for ThinkParQ’s BeeGFS and ArcaStream‘s PixStor file systems. Offering a combination of technology partners’ software with Dell EMC hardware, networking and support, based on engineered and tested designs, Ready Solutions for HPC Storage simplify and speed deployment and solutions management.

Ready Solutions for HPC BeeGFS Storage, with ThinkParQ’s software-defined parallel file system, speeds-up I/O-intensive workloads with the ability to scale from small clusters to enterprise systems on premises or in the cloud.

Additionally, Ready Solutions for HPC PixStor Storage offers a fast parallel file system, enabling data management at scale with the ability to perform archive and analytics in place. The solution also includes a validated, scalable design with object, tape and cloud tiering capabilities using PixStor Ngenea along with PowerEdge servers, PowerSwitch and Mellanox Technologies, Ltd. ‘s networking, PowerVault storage, supported by Dell deployment and support services.

With ready solutions, customers have seen improvements in performance and scale.
For example:

• Using the Ready Solutions for HPC BeeGFS Storage for its 2PB of all-NVMe storage, The Commonwealth Scientific and Industrial Research Organisation (CSIRO), Australia’s national science agency, stands as one of the largest NVMe storage systems in the world, currently ranked 8th on the IO-500 list, a benchmark that showcases the world’s fastest storage systems for HPC. ⁽³⁾

• At Imperial College London, the Ready Solutions for HPC PixStor Storage can simultaneously serve its existing 2,500 node HPC system and delivers over 20GB/s of throughput with no loss in interactive usage performance. ⁽⁴⁾

Powervault ME4084

The company also is introducing expanded capacity for PowerVault ME4, offering 16TB HDDs that allow customers to scale to 4PB in 15U rack space – a 25% improvement in density, allowing for more HPC storage capacity in a smaller space. PowerVault brings scale, bandwidth and built-in data protection to HPC configurations with management simplicity.

Simplifies path to AI with Ready Solutions and reference architectures
The firm is expanding its Ready Solutions for AI portfolio with a new validated design for the Domino Data Science Platform. Developed in collaboration with Domino Data Lab, the Dell EMC validated design enables data scientists to develop and deliver models faster while providing IT with a centralized, extensible platform spanning the entire data science lifecycle -accelerating ideation and deployment.

To simplify AI deployments, the company also is introducing 5 reference architectures for optimizing Dell EMC technologies with AI partners, such as those from DataRobot, Grid Dynamics, H20.ai, Iguazio, and Kubeflow on Red Hat OpenShift.

With these architectures, organizations can speed up deployment of AI solutions to modernize, automate and transform their data center using Dell EMC converged infrastructure, servers, storage and data protection technologies.

These architectures are optimized for Intel Xeon Scalable processors and enable organizations to speed up the deployment of AI solutions for training and inference to modernize, automate and transform their data centers using the firm’s converged infrastructure, servers, storage and data protection technologies.

400GbE networking switch for compute and storage intensive workloads

Joining the PowerSwitch Z-Series portfolio, the company is unveiling the PowerSwitch Z9332F-ON, a 400GbE open networking switch designed for performance workloads. As an open networking pioneer, the firm delivers on the promise of software-defined networking, making network operations more flexible, programmable and easier to manage.

The PowerSwitch Z9332F-ON is purpose-built for cloud service provider data center networks with intensive compute and storage traffic, such as HPC, AI and streaming video. This switch also delivers 4 times the throughput, double the price performance and near double the power efficiency of existing 100GbE platforms. ⁽⁵⁾

DSS 8440 server

Introduces Nividia GPU and Intel FPGA options across its server portfolio
The company is unveiling Nvidia T4 Tensor Core GPUs as an accelerator option for the DSS 8440 server. With up to 16 accelerators, this offers high capacity, high performance ML inference with energy efficiency (70W/GPU). This is designed for multi-tenancy environments that need to share ML resources among users or departments.

Introducing GPU and FPGA support for its PowerEdge servers, including:

• Nvidia Tesla V100S GPU with up to 25% more bandwidth to communicate directly with Mellanox IB interconnect and PowerEdge Express flash NVMe performance PCIe SSDs for faster data transfers.

• Nvidia RTX GPUs, designed to boost performance at a fraction of the cost, space and power requirements of a traditional render farm to produce quality content faster than before.

• Intel FPGA Programmable Acceleration Card D5005 in PowerEdge R740xd and R940xa servers to boost inferferencing, streaming analytics, video transcoding, and financial and genomic applications.

Availability:

• PowerVault ME4 with 16TB HDDs will be available in early 2020.

• Ready Solutions for AI – Design for Domino Data Science Platform will be globally available in December 2019.

• AI Reference architectures for Data Robot, Grid Dynamics, H20.ai and Iguazio have planned availability this quarter.

• Nvidia V100S GPUs will be available on PowerEdge servers in early 2020.

• Intel FPGA programmable acceleration card (PAC) D5005 will be available in early 2020.

• All other announced offerings and updates are available.

Harvey Newman, professor of Physics, Caltech, said: “We’ve been using Dell PowerSwitch Z Series products in both our network R&D and our HPC facility as part of the global Large Hadron Collider grid for years. We value the cost, performance, and reliability of the switches in HPC environments. Its SDN support has enabled us to develop new distributed system paradigms and intelligent networks to better serve the high energy physics and other data intensive science disciplines. The new Z9332F-ON is a highly capable 400GbE open networking switch that can deliver the fat pipes necessary to allow us to extend our SDN and system developments to a new scale, in order to meet the needs of our next data taking run at the LHC in 2021-3, and beyond.“

Nick Elprin, co-founder and CEO, Domino Data Labs, said: “It’s been deeply validating to see Dell extend a validated Domino data science platform design to clients who are on a journey to become model-driven. We really value the first-hand perspective and experience Dell Technologies has embedded into this solution and look forward to helping global enterprises deliver central, reproducible, and measurable data science results with the Dell EMC Ready Solutions for AI.“

⁽¹⁾ Hyperion Research, Worldwide HPC-based AI (AI) Market Forecast Update, 2018-2023, July 2019
⁽²⁾ Hyperion Research, Worldwide HPC Broader Market Forecast Update, 2018-2023, June 2019
⁽³⁾ IO-500 List, June 2019
⁽⁴⁾ ArcaStream PixStor User Case Study, Imperial College London: Protecting & Managing Data in World-Leading Research, 2019
⁽⁵) Based on internal Dell EMC analysis, October 1, 2019. Actual results will vary.