GPUs Add Up For ARM Chips In HPC

By Timothy Prickett Morgan

June 23, 2014

The first wave of credible 64-bit ARM processors are coming to market late this year or early next, and as is usually the case, the high-performance computing community is getting first crack at figuring out how these chips might be deployed to run various kinds of simulations more efficiently or cost effectively.

Applied Micro, which has first mover status in the 64-bit ARM server chip race with its X-Gene 1, is teaming up with Nvidia, maker of the Tesla GPU accelerators, at the International Super Computing conference in Leipzig, Germany to promote X-Gene and Tesla as the first of several dynamic duos. Three vendors – Cirrascale, E4 Computer Engineering, and Eurotech – are also previewing hybrid ARM-Tesla systems at the conference, and others will no doubt follow soon as more ARM chips come to market towards the end of this year and into early next year.

Given the ubiquity of Xeon processors in the supercomputing space, Nvidia has to integrate well with rival Intel’s Xeon processors and has to compete against the Xeon Phi parallel X86 coprocessors, too. But Nvidia, like many system buyers, wants a second or third option when it comes to processors, and that is why Nvidia was a founding member of the OpenPower Foundation, which seeks to establish multiple sources of IBM’s Power8 and follow-on processors and to link accelerators tightly to them. Nvidia is also waving the ARM banner high as well, and wants to be the accelerator of choice for ARMv8 platforms.

“GPUs make 64-bit ARM competitive in HPC on day one,” explains Ian Buck, general manager of GPU computing software at Nvidia. “We are clearly seeing viable and compelling ARM64 platforms coming online. It is obvious that there is excitement around ARM, and there are two reasons for that. One is that we haven’t had new, innovative CPUs for a while. Some of the ARM architectures are going up to 24 cores, and they are playing with what is on die, what is off, and Broadcom and Cavium come from the networking world and there are lots of networking angles they can play. The second reason for the excitement is choice. ARM represents choice, and a very diverse one.

nvidia-arm-hpc

While network devices like to have plenty of threads, the chips used in such gear are not generally equipped with lots of floating point math processing capability, says Buck. Nvidia, you can quickly guess, wants its Tesla to be the coprocessor of choice for 64-bit ARM platforms. Having created the CUDA programming environment, which supports 64-bit ARM chips starting with the 6.5 release, and a library of hundreds of third party simulation and analytics workloads to hybrid processor-GPU, Nvidia thinks it is well placed to help customers port their applications to ARM-Tesla hybrids.

“Based on our experience with ARM to date, the porting seems to go fairly quickly if you have well-structured code,” says Buck. “A lot of HPC codes have been around long enough that they don’t have a lot of intrinsics in there, the X86isms, and code seems to move fairly easily. If the code is already GPU-accelerated, then the performance just carries straight over. These ARM64 chips can drive full GPU performance.”

Applied Micro is going to have plenty of competition in the ARMv8 processor space, with AMD, Cavium, and Broadcom all putting forth very strong contenders to go up against the hegemony of Intel’s Xeon processors and its very credible defensive position with Atom chips for modest compute and low-power needs. Intel has a substantial lead in chip manufacturing processes – something between one and two nodes, depending on how you want to count it – and is behaving as if it has a bunch of AMDs on its heels. Never before in its history has Intel been so willing to tweak its processor designs to make them better fit the workloads of supercomputing and hyperscale customers alike, from adding special instructions to Xeons to baking special versions of the Xeons that run hotter or clock higher to actually welding an FPGA into a Xeon chip, as Intel last week announced it was going to do.

This newfound openness is one way Intel is going to counter the onslaught of different 64-bit ARM processors and the various ways their makers will accelerate workloads using GPUs, DSPs, FPGAs, and other specialized circuits. In effect, Intel is adopting the malleable approach of the ARM community to defend against ARM processors.

The initial X-Gene 1 processor from Applied Micro has been sampling since early 2013, and production wafers for the chip were started at the end of March and production chips are due around now. The X-Gene 1 chip is implemented in a 40 nanometer process at Taiwan Semiconductor Manufacturing Corp; it has eight custom ARMv8 cores, designed by Applied Micro itself, on each system-on-chip. The cores on the X-Gene 1 run at 2.4 GHz, and Sanchayan Sinha, senior product manager, tells HPCwire that in terms of single-threaded performance, the X-Gene 1 has about the same level of oomph as a four-core “Haswell” Xeon E3 and about the same memory bandwidth as a “Sandy Bridge” Xeon E5.

Sinha stressed that these were very rough comparisons and that real benchmarks would eventually result in harder figures than these approximations. That is, in fact, what the development systems being shown off at ISC’14 are all about. The company is working with server partners to run the High Performance Conjugate Gradients (HPCG) benchmark, which is being proposed as a follow-on to the more widely used Linpack parallel Fortran matrix math test, on X-Gene 1 systems. Sinha says that Applied Micro and Nvidia will be able to show that an X-Gene 1 plus a Tesla K20 coprocessor will be equivalent to an X86 processor plus the same Tesla K20 floating point motor.

x-gene-1-block

The X-Gene 2 chip is a rev on the initial design and also includes eight ARM cores, but it is implemented in a 28 nanometer process at TSMC. This shrink of the process will allow Applied Micro to crank up the clock speed and add more features to its SoC. One interesting feature that the company has divulged it will add to the X-Gene 2 is support for Remote Direct Memory Access (RDMA) on the network ports on the chip. Specifically, the Ethernet ports on the chip will be able to run RDMA over Converged Ethernet (RoCE), which brings the low-latency access of InfiniBand to the Ethernet protocol. This will make the X-Gene 2 chip not only suitable for HPC workloads that are latency sensitive, but also for database, storage, and transaction processing workloads in enterprise datacenters that also like low latency.

Further out beyond this, Applied Micro has teamed up with TSMC to use its 16 nanometer FinFET 3D transistor process to create X-Gene 3. Little is known about this processor except that it will have at least 16 cores on the SoC.

This early revs of the X-Gene 1 were put on development boards called “Mustang” internally by Applied Micro and known as the X-Gene XC-1 outside of the company. The ARM-based HPC systems that are being previewed by Cirrascale and E4 Computer Engineering are based on production-grade X-Gene 1 chips and the Mustang boards.

The Cirrascale development machine puts two Mustang boards and two Tesla K20 or K20X GPU accelerators in a compact 1U server chassis:

cirrascale-x-gene

This machine is called the RM1905D in the Cirrascale product catalog, and like other Mustang board it supports a maximum of 64 GB of memory for each X-Gene 1 chip across the processor’s two memory slots. The system has four Ethernet ports: three for data and one for system management. Two of the ports for data exchange run at 1 Gb/sec and the remaining one runs at 10 Gb/sec; the management port runs at 1 Gb/sec. The Mustang board has one PCI-Express 3.0 x8 slot, which is used to link the processor to the Tesla GPU, and the chassis has room to plug in a single SATA-2 drive (a 6 Gb/sec link). Each node in the chassis has a 400 watt power supply.

The feeds and speeds of E4 Computer Engineering’s EK003 were not available at press time, but Nvidia tells HPCwire that the machine will include two X-Gene 1 system boards in a 3U enclosure that has two Tesla K20 GPU coprocessors, and that the development machine will be aimed at seismic, signal and image processing, video analytics, track analysis, Web applications, and MapReduce workloads.

Cirrascale and E4 Computer Engineering plan to ship their development machines in July, according to Nvidia.

Eurotech has a custom motherboard design using the X-Gene 1 chip that has main memory soldered onto the board to give it a very low profile and therefore high density for its ARM-based Aurora system. The compute elements in this new Aurora machine are based on what the company calls its “brick technology,” and will employ direct hot-water cooling of the components in the brick. It will include a combination of ARM processors and Tesla coprocessors. Further details for this Eurotech Aurora system were not yet available at press time, but we will hunt them down. The company expects to ship production machines later this year.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industy updates delivered to you every week!

UCSD, AIST Forge Tighter Alliance with AI-Focused MOU

January 18, 2018

The rich history of collaboration between UC San Diego and AIST in Japan is getting richer. The organizations entered into a five-year memorandum of understanding on January 10. The MOU represents the continuation of a 1 Read more…

By Tiffany Trader

New Blueprint for Converging HPC, Big Data

January 18, 2018

After five annual workshops on Big Data and Extreme-Scale Computing (BDEC), a group of international HPC heavyweights including Jack Dongarra (University of Tennessee), Satoshi Matsuoka (Tokyo Institute of Technology), Read more…

By John Russell

Researchers Measure Impact of ‘Meltdown’ and ‘Spectre’ Patches on HPC Workloads

January 17, 2018

Computer scientists from the Center for Computational Research, State University of New York (SUNY), University at Buffalo have examined the effect of Meltdown and Spectre security updates on the performance of popular H Read more…

By Tiffany Trader

HPE Extreme Performance Solutions

HPE and NREL Take Steps to Create a Sustainable, Energy-Efficient Data Center with an H2 Fuel Cell

As enterprises attempt to manage rising volumes of data, unplanned data center outages are becoming more common and more expensive. As the cost of downtime rises, enterprises lose out on productivity and valuable competitive advantage without access to their critical data. Read more…

Fostering Lustre Advancement Through Development and Contributions

January 17, 2018

Six months after organizational changes at Intel's High Performance Data (HPDD) division, most in the Lustre community have shed any initial apprehension around the potential changes that could affect or disrupt Lustre Read more…

By Carlos Aoki Thomaz

UCSD, AIST Forge Tighter Alliance with AI-Focused MOU

January 18, 2018

The rich history of collaboration between UC San Diego and AIST in Japan is getting richer. The organizations entered into a five-year memorandum of understandi Read more…

By Tiffany Trader

New Blueprint for Converging HPC, Big Data

January 18, 2018

After five annual workshops on Big Data and Extreme-Scale Computing (BDEC), a group of international HPC heavyweights including Jack Dongarra (University of Te Read more…

By John Russell

Researchers Measure Impact of ‘Meltdown’ and ‘Spectre’ Patches on HPC Workloads

January 17, 2018

Computer scientists from the Center for Computational Research, State University of New York (SUNY), University at Buffalo have examined the effect of Meltdown Read more…

By Tiffany Trader

Fostering Lustre Advancement Through Development and Contributions

January 17, 2018

Six months after organizational changes at Intel's High Performance Data (HPDD) division, most in the Lustre community have shed any initial apprehension aroun Read more…

By Carlos Aoki Thomaz

When the Chips Are Down

January 11, 2018

In the last article, "The High Stakes Semiconductor Game that Drives HPC Diversity," I alluded to the challenges facing the semiconductor industry and how that may impact the evolution of HPC systems over the next few years. I thought I’d lift the covers a little and look at some of the commercial challenges that impact the component technology we use in HPC. Read more…

By Dairsie Latimer

How Meltdown and Spectre Patches Will Affect HPC Workloads

January 10, 2018

There have been claims that the fixes for the Meltdown and Spectre security vulnerabilities, named the KPTI (aka KAISER) patches, are going to affect applicatio Read more…

By Rosemary Francis

Momentum Builds for US Exascale

January 9, 2018

2018 looks to be a great year for the U.S. exascale program. The last several months of 2017 revealed a number of important developments that help put the U.S. Read more…

By Alex R. Larzelere

ANL’s Rick Stevens on CANDLE, ARM, Quantum, and More

January 8, 2018

Late last year HPCwire caught up with Rick Stevens, associate laboratory director for computing, environment and life Sciences at Argonne National Laboratory, f Read more…

By John Russell

Inventor Claims to Have Solved Floating Point Error Problem

January 17, 2018

"The decades-old floating point error problem has been solved," proclaims a press release from inventor Alan Jorgensen. The computer scientist has filed for and Read more…

By Tiffany Trader

US Coalesces Plans for First Exascale Supercomputer: Aurora in 2021

September 27, 2017

At the Advanced Scientific Computing Advisory Committee (ASCAC) meeting, in Arlington, Va., yesterday (Sept. 26), it was revealed that the "Aurora" supercompute Read more…

By Tiffany Trader

Japan Unveils Quantum Neural Network

November 22, 2017

The U.S. and China are leading the race toward productive quantum computing, but it's early enough that ultimate leadership is still something of an open questi Read more…

By Tiffany Trader

AMD Showcases Growing Portfolio of EPYC and Radeon-based Systems at SC17

November 13, 2017

AMD’s charge back into HPC and the datacenter is on full display at SC17. Having launched the EPYC processor line in June along with its MI25 GPU the focus he Read more…

By John Russell

Nvidia Responds to Google TPU Benchmarking

April 10, 2017

Nvidia highlights strengths of its newest GPU silicon in response to Google's report on the performance and energy advantages of its custom tensor processor. Read more…

By Tiffany Trader

IBM Begins Power9 Rollout with Backing from DOE, Google

December 6, 2017

After over a year of buildup, IBM is unveiling its first Power9 system based on the same architecture as the Department of Energy CORAL supercomputers, Summit a Read more…

By Tiffany Trader

Fast Forward: Five HPC Predictions for 2018

December 21, 2017

What’s on your list of high (and low) lights for 2017? Volta 100’s arrival on the heels of the P100? Appearance, albeit late in the year, of IBM’s Power9? Read more…

By John Russell

GlobalFoundries Puts Wind in AMD’s Sails with 12nm FinFET

September 24, 2017

From its annual tech conference last week (Sept. 20), where GlobalFoundries welcomed more than 600 semiconductor professionals (reaching the Santa Clara venue Read more…

By Tiffany Trader

Leading Solution Providers

Chip Flaws ‘Meltdown’ and ‘Spectre’ Loom Large

January 4, 2018

The HPC and wider tech community have been abuzz this week over the discovery of critical design flaws that impact virtually all contemporary microprocessors. T Read more…

By Tiffany Trader

Perspective: What Really Happened at SC17?

November 22, 2017

SC is over. Now comes the myriad of follow-ups. Inboxes are filled with templated emails from vendors and other exhibitors hoping to win a place in the post-SC thinking of booth visitors. Attendees of tutorials, workshops and other technical sessions will be inundated with requests for feedback. Read more…

By Andrew Jones

Tensors Come of Age: Why the AI Revolution Will Help HPC

November 13, 2017

Thirty years ago, parallel computing was coming of age. A bitter battle began between stalwart vector computing supporters and advocates of various approaches to parallel computing. IBM skeptic Alan Karp, reacting to announcements of nCUBE’s 1024-microprocessor system and Thinking Machines’ 65,536-element array, made a public $100 wager that no one could get a parallel speedup of over 200 on real HPC workloads. Read more…

By John Gustafson & Lenore Mullin

Delays, Smoke, Records & Markets – A Candid Conversation with Cray CEO Peter Ungaro

October 5, 2017

Earlier this month, Tom Tabor, publisher of HPCwire and I had a very personal conversation with Cray CEO Peter Ungaro. Cray has been on something of a Cinderell Read more…

By Tiffany Trader & Tom Tabor

Flipping the Flops and Reading the Top500 Tea Leaves

November 13, 2017

The 50th edition of the Top500 list, the biannual publication of the world’s fastest supercomputers based on public Linpack benchmarking results, was released Read more…

By Tiffany Trader

GlobalFoundries, Ayar Labs Team Up to Commercialize Optical I/O

December 4, 2017

GlobalFoundries (GF) and Ayar Labs, a startup focused on using light, instead of electricity, to transfer data between chips, today announced they've entered in Read more…

By Tiffany Trader

How Meltdown and Spectre Patches Will Affect HPC Workloads

January 10, 2018

There have been claims that the fixes for the Meltdown and Spectre security vulnerabilities, named the KPTI (aka KAISER) patches, are going to affect applicatio Read more…

By Rosemary Francis

HPC Chips – A Veritable Smorgasbord?

October 10, 2017

For the first time since AMD's ill-fated launch of Bulldozer the answer to the question, 'Which CPU will be in my next HPC system?' doesn't have to be 'Whichever variety of Intel Xeon E5 they are selling when we procure'. Read more…

By Dairsie Latimer

  • arrow
  • Click Here for More Headlines
  • arrow
Share This