Why -- and how -- high-performance computing technology is coming to your data center

Presented by AMD

Increasing pressure on performance has been a fact of life in the data center environment for several years now. Compute intensive workloads have become more entrenched and more demanding for data centers to handle. They are built on complex, constantly evolving models that ask a lot of today’s infrastructures. Servers, storage capacity, memory, bandwidth — all must be at peak performance, all the time, while also functioning with minimal power consumption and within the tightest footprints. IT leaders are being faced with the requirements to provide increasing levels of performance, but with tighter resources.

To power the computer-intensive needs of AI, 94.4% of companies with more than 2,500 employees said optimized server solutions were required.

The growing demand from business leaders for these high-performance data centers was recently revealed in an AMD-sponsored IDC white paper, High-Performance Computing Drives Critical New Capabilities for Mainstream Organizations. IDC’s survey of IT decision-makers at companies with over 2,500 employees found that 94.4% of respondents required optimized server solutions for performance-intensive computing applications such as artificial intelligence.

High-performance computing in the enterprise

Few organizations are unaffected by the need for high-performance computing. Not long ago, conventional thinking was that high-performance computing was only required for exceptionally data-intensive applications within select industries — aerospace, oil and gas, and pharmaceuticals, for example, in addition to supercomputing centers dedicated to solving large, complex problems.

This is no longer the case. As data volumes have exploded, many organizations are tapping into these technology and techniques to perform essential functions. In a relatively short timeframe, they’ve gone from believing they would never need anything beyond routine compute performance capabilities, to depending on high-performance computing to fuel their business success. Examples include keeping application logins running smoothly, running algorithms that keep systems secure, or even monitoring retail environments with computer vision to prevent shoplifting.

No longer the domain of a few select fields, industries from financial trading to the advertising ecosystem rely on large-scale mathematically intensive computations.

In conjunction with AI and data analytics, high-performance computing is powering entire industries that depend for their existence on performing large-scale, mathematically intensive computations for a variety of needs, including faster business insights and results to drive improved decision-making. Financial trading is increasingly based on algorithms and machine learning (ML), and real-time data on customer interactions drives online advertising. Whenever businesses design and simulate products, the computer-aided engineering applications they use also need high-performance, high-efficiency computing.

New workloads call for new solutions

Depending on the organization and its IT requirements, high performance in the data center can be required to perform and support certain key functions:

Online transaction processing (OLTP) is at the core of many enterprise applications. Fast application performance — with more customer transactions per server — ultimately delivers improved sales.

Query performance is essential to tapping into the power of relational database management systems for analytics and decision support. Faster time-to-insights means the organization can change direction quickly to adjust to shifting market and customer requirements.

Virtualization performance is another critical factor in HPC infrastructure design. By running on a layer of virtualization, applications make more effective use of hardware resources.

Nothing is more central to driving high data center performance than processor choice.

To support IT organizations as they grapple with managing and processing massive amounts of data to accelerate business results, technology providers are bringing innovative products and services to market. But nothing is more central to driving high data center performance than processor choice.

Taking transaction processing as an example, a server powered by a 2P (dual processor) 4th Gen AMD EPYC 9654 processor delivers approximately a 2.71x performance advantage compared to the Intel Xeon 8380 2P solution.¹ On query performance, a 2P AMD EPYC 9654 processor-driven server delivers about a 2.7x median performance improvement over the Intel 8380 2P offering.²

Similar advantages are apparent in virtualization performance. For example, in a head-to-head between two 2P servers, an EPYC 9654 processor solution outscored the Intel Xeon 8480H-based solution by 1.7x in VMmark matched pair benchmark, an industry standard measure of performance, scalability, and power consumption of virtualization platforms.³

How Emirates NBD modernized for HPC

High-performance solutions like these are in demand for companies seeking to modernize their data centers, build out private or hybrid clouds or drive ultimate performance even at the edge.

Emirates NBD saw increases in speed from 42% to 51%. Processing terabytes of logs daily, they went from waiting 30 seconds or minutes, to almost instantaneous processing.

A case in point is Dubai-based Emirates NBD Bank. In 2018, as part of a thorough overhaul of its IT infrastructure, the bank subjected a number of proposals to intense evaluation. Ali Rey, senior vice president for Emirates NBD’s technology platform, recalls: “We tested our database performance, our core banking APIs, the memory bandwidth and how the CPUs accessed the memory. And we did a lot of web-based user testing. Eventually, we went with building out a private cloud environment powered by 3rd Gen AMD EPYC CPUs, which we found to be on average 42% faster than the alternative solution — 51% faster, in some cases.”

The servers powered by the AMD CPUs are also more effective at the ongoing, processor-intensive maintenance routines all businesses depend on. According to Rey, “We process terabytes of logs every day. Before we had to wait over 30 seconds to minutes, but now they’re almost instantaneous.” He adds that the multicore processors are hugely more efficient, delivering a lot more processing for the same infrastructure density with fewer servers. “There aren’t many banks that can match our level of efficiency.”

Efficient data center computing

Many enterprises have concerns that achieving results like those at Emirates NBD will require larger data centers — increasing costs and energy usage. However, today’s newer CPUs can allow IT leaders and enterprises to meet their performance needs with even greater efficiency.

Surprising many, larger data centers are not a requirement — rather, today’s newer CPUs achieve the greater efficiency IT leaders are after.

For example, the server capacity required to run 2,000 virtual machines would call for 17 2P Intel Platinum 8490H-based servers, but only 11 servers based on a 2P 96-core 4th Gen AMD EPYC processors. That’s the same amount of work with 35% fewer servers and an estimated 36% less power consumption annually.⁴

Efficiency gains at this scale are vital to enabling enterprises to achieve their data center performance goals, while also meeting business goals such as lowering energy costs and advancing broader corporate sustainability initiatives. Taking advantage of the latest CPUs now in market can allow IT leaders to meet the performance needs their organizations and customers require, all while achieving greater cost and energy efficiency.

Ravi Kuppuswamy is AMD’s corporate vice president for the Server Solutions Group.

Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact sales@venturebeat.com.