In recent years, traditional supercomputing competitors, like SGI (now part of HPE), Cray, and IBM, have seen mainstream server HPC brands, like Dell EMC, HPE, and Lenovo, gain end-user attention. For the past few years the International Supercomputing Conference (ISC), a leading conference for high-performance computing (HPC) held in Frankfurt, Germany last week, has held a vendor showdown: an audience-judged competition among ISC’s Platinum, Gold and Silver sponsors.
Audience judging matters at ISC, because its audience is so highly qualified. [You can read the details for 2016’s event here. I assume this year was similar, but ISC had not yet posted the 2017 split before this was published.] ISC’s vendor showdown is a timed event; each vendor has a few minutes to present their strategy, product or research developments using only three slides, and must then answer questions from a panel of moderators.
This year, the 22 participating companies were split evenly into two tracks. Dell EMC participated in track 2 (listed here in the order they presented):
Dell EMC’s Ed Turkel gave the winning presentation for track 2. Dell EMC has been experimenting and innovating around the company’s mission to “democratize HPC” (remember that the PC mission belongs to Dell now, a separate corporate entity from Dell EMC). Democratizing HPC means making HPC accessible to small and medium businesses to accelerate science, engineering, and analytics applications. It is doing this by working with partners to advance the technology base and by optimizing HPC solutions, via their Dell EMC HPC Innovation lab.
Dell EMC’s Ed Turkel posing with the vendor showdown trophy in front of a rack full of new... [+]
Dell EMC is working with subject matter experts and dedicated vertical market sales and solutions architects to understand specific workloads for their target verticals. The result is that Dell EMC’s HPC Systems for Research, Manufacturing, Life sciences, Lustre, and Network File System (NFS) enable customers to size and optimize HPC clusters for those workloads.
Dell EMC Customer examples
Late last year, MIT Lincoln Laboratory Supercomputing Center (LLSC) finished installing its Dell EMC-based “TX-Green” cluster, which ranks 125 on the June 2017 supercomputer Top500 list with over one peta-FLOP (floating point operations per second) performance. LLSC’s TX-Green cluster is used for research into many cutting-edge applications, including cyber security, Machine Learning (ML) and bioinformatics.
Mastercard has built a secure, Payment Card Industry (PCI) certified Hadoop cluster based on Dell EMC systems. Mastercard uses ML to examine 160 million credit card transactions per hour from 2.2 billion credit cards in use, using only milliseconds per transaction to better identify fraud. Its fraud detection ML system uses supervised learning to look for established fraud patterns and unsupervised learning to identify emerging fraud patterns in real-time.
Dell EMC Expands Partner Ecosystem
Last week at ISC, Dell EMC announced two new partnerships – with CoolIT, to bring water cooling into mainstream and lower-cost HPC market segments, and a deeper, expanded relationship with NVIDIA for joint product development around Dell EMC’s HPC infrastructure and NVIDIA’s GPU-accelerated solutions for HPC, data analytics, ML and Artificial Intelligence (AI).
I think that one of the most telling aspects of Dell EMC’s HPC growth is that it does not have an exascale research and development (R&D) program. [Here is a good description of exascale.] Its growth in HPC is possible because several of its suppliers and partners are investing in exascale R&D. For example, NVIDIA, AMD, and Intel recently received U.S. Department of Energy (DoE) exascale research contracts (the DoE did not publish how much each company received from the $258 million total award).
Dell EMC has historically partnered with Bright Computing to offer software development and runtime environments for HPC clusters, and that is also true for ML and Dell EMC’s new NVIDIA partnership. Bright Cluster Manager Advanced Edition offers NVIDIA CUDA & OpenCL support, along with NVIDIA Tesla GPU and Intel Xeon Phi management and monitoring.
As Dell EMC moves forward, I think it will have to move beyond partnerships to a richer mix of R&D and forward-looking mergers and acquisitions (M&A) activity. Dell Technologies Capital was unveiled in May to invest in emerging technology companies, and has already invested in ML/AI and genome analytics acceleration companies (Graphcore and Edico Genome, respectively). However, that does not address internal R&D as well as outright M&A to create and secure access to high value intellectual property.
Dell EMC HPC Innovation Lab and Product Updates
Another measure of Dell EMC’s commitment to HPC is that its own internal HPC Innovation Lab’s Zenith cluster debuted at 372 in the Top500 list in November 2016 with 13,824 Intel Xeon cores, and has since been upgraded to 19,920 Intel Xeon cores and ranks 373 in the June 2017 list. Dell EMC added 44% more cores to the Zenith cluster in six months to lose only one place out of 500. One place out of 500 is effectively in the noise; the Zenith cluster is keeping pace with Top500 supercomputer capabilities.
Dell EMC’s HPC Innovation Lab also includes a 32-node Skylake cluster with Intel Omni-path networking for testing and evaluation, PowerEdge C4130 servers with Nvidia GPUs and Mellanox InfiniBand network adapters, as well as mixed Intel Xeon and Intel Xeon Phi C6320p systems.
From its Triton rack-level water cooling architecture to its PowerEdge C4130 1U/2S system that can house up to four full-sized, double-width PCIe accelerator cards, Dell EMC has shown a practical approach to incremental innovation.
The C4130 comes in two variants: the default houses PCIe add-in-boards and a new variant that houses four NVIDIA SMX2 modules connected via NVLINK. The PCIe variant supports multiple PCIe root topologies using optional PLX switches. The C4130 can also support NVIDIA’s GPUDirect heterogeneous memory architecture using either InfiniBand or Ethernet.
Dell EMC is upgrading its Broadwell-based PowerEdge C6320 to the Skylake-based C6420, which can be optionally cooled using CoolIT liquid cooling technology. It will keep the Xeon Phi “Knights Landing” based C6320p in the product line-up, presumably until Intel’s next generation “Knights Mill” is available.
In addition, Dell EMC’s PowerEdge R740 can host three 300W or six 150W accelerator add-in boards, and is intended for ML, AI, and virtual desktop infrastructure (VDI) applications. The R740 can also substitute NVDIMM memory for DRAM based memory in half of its 24 memory slots, useful for accelerating in-memory analytics.
While the C4130, C6420 and R740 are keeping up with the leading edge of Intel-based server architecture, I am looking forward to seeing the results of Dell EMC’s announced support for AMD’s recently launched EPYC processor. Dell EMC needs to differentiate from the rest of the Intel-based HPC hardware solutions pack. AMD EPYC may give Dell EMC the opportunity to do so.
Congratulations to Dell EMC on winning its ISC vendor showdown track! This high level of acknowledgement from ISC’s highly qualified attendee base is a clear sign that Dell EMC’s strategy and velocity are being recognized by a broad base of customers and researchers.
-- The author and members of the TIRIAS Research staff do not hold equity positions in any of the companies mentioned. TIRIAS Research tracks and consults for companies throughout the electronics ecosystem from semiconductors to systems and sensors to the cloud, including Dell EMC, AMD and NVIDIA.
