The Next Platform

Engineering Code Scales Across 200,000 Cores on Cray Super

Teams at Saudi Aramco using the Shaheen II at King Abdullah University of Science and Technology (KAUST) supercomputer have managed to scale ANSYS Fluent across 200,000 cores, marking top-end scaling for the commercial engineering code.

The news last year of a code scalability effort that topped out at 36,000 cores on the Blue Waters machine at the National Center for Supercomputing Applications (NCSA) was impressive. That was big news for ANSYS and NCSA, but also a major milestone for Cray. Just as Blue Waters is a Cray system, albeit one at the outer reaches of its lifespan (it was installed …
Read more
The Trials And Tribulations Of IBM Systems

IBM is a bit of an enigma these days. It has the art – some would say black magic – of financial engineering down pat, and its system engineering is still quite good. Big Blue talks about all of the right things for modern computing platforms, although it speaks a slightly different dialect because the company still thinks that it is the one setting the pace, and therefore coining the terms, rather than chasing markets that others are blazing. And it just can’t seem to grow revenues, even after tens of billions of dollars in acquisitions and internal investments over …
Read more
How Google Wants To Rewire The Internet

When all of your business is driven by end users coming to use your applications over the Internet, the network is arguably the most critical part of the infrastructure. That is why search engine and ad serving giant Google, which has expanded out to media serving, hosted enterprise applications, and cloud computing, has put a tremendous amount of investment into creating its own network stack.

But running a fast, efficient, hyperscale network for internal datacenters is not sufficient for a good user experience, and that is why Google has created a software defined networking stack to do routing over the …
Read more
The Golden Grail: Automatic Distributed Hyperparameter Tuning

While it might not be an exciting problem front and center of AI conversations, the issue of efficient hyperparameter tuning for neural network training is a tough one. There are some options that aim to automate this process but for most users, this is a cumbersome area—and one that can lead to bad performance when not done properly.

The problem with coming up with automatic tools for tuning is that many machine learning workloads are dependent on the dataset and the conditions of the problem being solved. For instance, some users might prefer less accuracy over a speedup or efficiency …
Read more
The System Bottleneck Shifts To PCI-Express

No matter what, system architects are always going to have to contend with one – and possibly more – bottlenecks when they design the machines that store and crunch the data that makes the world go around. These days, there is plenty of compute at their disposal, a reasonable amount of main memory to hang off of it, and both Ethernet and InfiniBand are on the cusp of 200 Gb/sec of performance and not too far away from 400 Gb/sec and even higher bandwidths.

Now, it looks like the peripheral bus based on the PCI-Express protocol is becoming the bottleneck, …
Read more
Technology Requirements for Deep and Machine Learning

Having been at the forefront of machine learning since the 1980s when I was a staff scientist in the Theoretical Division at Los Alamos performing basic research on machine learning (and later applying it in many areas including co-founding a machine-learning based drug discovery company), I was lucky enough to participate in the creation and subsequently to observe first-hand the process by which the field of machine-learning grew to become a ‘bandwagon’ that eventually imploded due to misconceptions about the technology and what it could accomplish.

Fueled by across-the-board technology advances including algorithmic developments, machine learning has again become a …
Read more
The New Server Economies Of Scale For AMD

In the first story of this series, we discussed the Infinity fabric that is at the heart of the new “Naples” Epyc processor from AMD, and how this modified and extended HyperTransport interconnect glues together the cores, dies, and sockets based on Eypc processors into a unified system.

In this follow-on story, we will expand out from the Epyc processor design to the basic feeds and speeds of the system components based on this chip and then take a look at some of the systems that AMD and its partners were showing off at the Epyc launch a few …
Read more
The Convergence Or Divergence Of HPC And AI Iron

Based on datacenter practices of the past two decades, it is a matter of faith that it is always better to run a large number of applications on a given set of generic infrastructure than it is to have highly tuned machines running specific workloads. Siloed applications on separate machines are a thing of the past. However, depending on how Moore’s Law progresses (or doesn’t) and how the software stacks shake out for various workloads, organizations might be running applications on systems with very different architectures, either in a siloed, standalone fashion or across a complex workflow that links the …
Read more
The Heart Of AMD’s Epyc Comeback Is Infinity Fabric

At AMD’s Epyc launch few weeks ago, Lisa Su, Mark Papermaster, and the rest of the AMD Epyc team hammered home that AMD designed its new Zen processor core for servers first. This server-first approach has implications for performance, performance per watt, and cost perspectives in both datacenter and consumer markets.

AMD designed Epyc as a modular architecture around its “Zeppelin” processor die with its eight “Zen” architecture cores. To allow multi-die scalability, AMD first reworked its HyperTransport socket-to-socket system I/O architecture for use on a chip, across a multi-chip module (MCM), and for inter-socket connectivity. AMD has named this …
Read more
Securing The HPC Infrastructure

In the world of high performance computing (HPC), the most popular buzzwords include speed, performance, durability, and scalability. Security is one aspect of HPC that is not often discussed, or else seems to be relatively low on the list of priorities when organizations begin building out their infrastructures to support a demanding new applications, whether for oil and gas exploration, machine learning, simulations, or visualization of complex datasets.

While IT security is paramount for businesses in the digital age, HPC systems typically do not encounter the same risks as public-facing infrastructures – in the same way as a web server …
Read more
High Expectations for Low Precision at CERN

The last couple of years has seen a steady drumbeat for the use of low precision in growing numbers of workloads driven in large part by the rise of machine learning and deep learning applications and the ongoing desire to cut back on the amount of power consumed.

The interest in low precision is rippling through the high-performance computing (HPC) field, spanning companies that are running applications sets to the tech vendors that are creating the systems and components on which the work is done.

The Next Platform has kept a steady eye on the developments the deep-learning and machine-learning …
Read more
The X86 Battle Lines Drawn With Intel’s Skylake Launch

At long last, Intel’s “Skylake” converged Xeon server processors are entering the field, and the competition with AMD’s “Naples” Epyc X86 alternatives can begin and the ARM server chips from Applied Micro, Cavium, and Qualcomm and the Power9 chip from IBM know exactly what they are aiming at.

It is a good time to be negotiating with a chip maker for compute power.

The Skylake chips, which are formally known as the Xeon Scalable Processor family, are the result of the convergence of the workhorse Xeon E5 family of chips for two-socket and four-socket servers with the higher-end Xeon E7 …
Read more
China Tunes Neural Networks for Custom Supercomputer Chip

Supercomputing centers around the world are preparing their next generation architectural approaches for the insertion of AI into scientific workflows. For some, this means retooling around an existing architecture to make capability of double-duty for both HPC and AI.

Teams in China working on the top performing supercomputer in the world, the Sunway TaihuLight machine with its custom processor, have shown that their optimizations for theSW26010 architecture on deep learning models have yielded a 1.91-9.75X speedup over a GPU accelerated model using the Nvidia Tesla K40m in a test convolutional neural network run with over 100 parameter configurations.

Efforts on …
Read more
OpenPower, Efficiency Tweaks Define Europe’s DAVIDE Supercomputer

When talking about the future of supercomputers and high-performance computing, the focus tends to fall on the ongoing and high-profile competition between the United States with its slowly eroding place as the kingpin in the industry and China and the tens of billions of dollars that the government has invested in recent years to rapidly expand the reach of the country’s tech community and the use of home-grown technologies in massive new systems.

Both trends were on display at the recent International Supercomputing Conference in Frankfurt, Germany, where China not only continued to hold the top two spots on the …
Read more
Ethernet Getting Back On The Moore’s Law Track

It would be ideal if we lived in a universe where it was possible to increase the capacity of compute, storage, and networking at the same pace so as to keep all three elements expanding in balance. The irony is that over the past two decades, when the industry needed for networking to advance the most, Ethernet got a little stuck in the mud.

But Ethernet has pulls out of its boots and left them in the swamp and is back to being barefoot again on much more solid ground where it can run faster. The move from 10 Gb/sec …
Read more
Parameter Encoding on FPGAs Boosts Neural Network Efficiency

The key to creating more efficient neural network models is rooted in trimming and refining the many parameters in deep learning models without losing accuracy. Much of this work is happening on the software side, but devices like FPGAs that can be tuned for trimmed parameters are offering promising early results for implementation.

A team from UC San Diego has created a reconfigurable clustering approach to deep neural networks that encodes the parameters the network according the accuracy requirements and limitations of the platform—which are often bound by memory access bandwidth. Encoding the trimmed parameters in an FPGA resulted in …
Read more
Using The Network To Break Down Server Silos

Virtual machines and virtual network functions, or VMs and VNFs for short, are the standard compute units in modern enterprise, cloud, and telecommunications datacenters. But varying VM and VNF resource needs as well as networking and security requirements often force IT departments to manage servers in separate silos, each with their own respective capabilities.

For example, some VMs or VNFs may require a moderate number of vCPU cores and lower I/O bandwidth, while VMs and VNFs associated with real-time voice and video, IoT, and telco applications require a moderate-to-high number of vCPU cores, rich networking services, and high I/O bandwidth, …
Read more
A High Performance Hiatus

The Next Platform will be on summer holiday for the coming week.

We will return to our normal publishing schedule on Monday, July 10^th.

For our U.S. readership, we hope you have an excellent Independence Day holiday week and for our many readers outside the states, here’s hoping you find time to relax and enjoy great weather.

Thanks as always for reading, and for our sponsors, we appreciate your support.

Cheers,

Timothy Prickett Morgan, Nicole Hemsoth

Co-Founders, Co-Editors, The Next Platform …
Read more
InfiniBand And Proprietary Networks Still Rule Real HPC

With the network comprising as much as a quarter of the cost of a high performance computing system and being absolutely central to the performance of applications running on parallel systems, it is fair to say that the choice of network is at least as important as the choice of compute engine and storage hierarchy. That’s why we like to take a deep dive into the networking trends present in each iteration of the Top 500 supercomputer rankings as they come out.

It has been a long time since the Top 500 gave a snapshot of pure HPC centers that …
Read more
Fix Your NAS With Metadata

Enterprises are purchasing storage by the truckload to support an explosion of data in the datacenter. IDC reports that in the first quarter of 2017, total capacity shipments were up 41.4 percent year-over-year and reached 50.1 exabytes of storage capacity shipped. As IT departments continue to increase their spending on capacity, few realize that their existing storage is a pile of gold that can be fully utilized once enterprises overcome the inefficiencies created by storage silos.

A metadata engine can virtualize the view of data for an application by separating the data (physical) path from the metadata (logical) path. This …
Read more

Previous Articles