CPU-based Visualization Positions for Exascale Supercomputing

By Jim Jeffers, Principal Engineer and Engineering Leader, Intel

March 16, 2017

Editor’s Note: In this contributed perspective piece, Intel’s Jim Jeffers makes the case that CPU-based visualization is now widely adopted and as such is no longer a contrarian view, but is rather an exascale requirement. This article is a follow-on to a 2015 contribution from Jeffers and traces the progress for CPU-based software-defined visualization from that time.

Since our first formal product releases of OSPRay and OpenSWR libraries in 2016, CPU-based Software Defined Visualization (SDVis) has achieved wide-spread adoption. This rapid uptake is the result of two factors: (1) the general availability of highly-optimized CPU-based rendering software such as the open-source OSPRay ray tracing library and the high performance OpenSWR raster library in Mesa3d  integrated into popular visualization tools like Kitware’s Paraview and VTK, as well as the community tool, VisIt; and (2) SDVis filling the big data visualization community need for software that uses runtime visualization algorithms that can handle giga-scale and larger data.

These technologies aim to enable production visualization at scale on high performance computing resources, including supercomputers at Argonne National Laboratory, Los Alamos National Laboratory, the Texas Advanced Computing Center and many other facilities.

Award winning results, such as the Best Visualization and Data Analytics Showcase award won by the Los Alamos’ Data Science at Scale Team at Supercomputing 2016, highlight the fact that CPU-based rendering is now at the forefront of visualization technology. The LANL team’s award winning asteroid impact visualization is featured as an LANL newsroom picture of the week.

Figure 1: One image from the LANL asteroid impact video (Source: LANL)

Dr. Aaron Knoll (Research Scientist, Scientific Computing and Imaging Institute at the University of Utah) explains that the key change from last year lies in how much OSPRay and other SDVis CPU-based visualization libraries are now being used. “2016 is the year OSPRay became used in practice and production,” he said.

This trend has occurred throughout the scientific community. For example, four out of six finalists at Supercomputing 2016 used OSPRay and/or OpenSWR for their CPU-based visualizations. Of the remaining two finalists, one expressed interest in VMD rendering using OSPRay (now supported by that package), and the other used purely information visualization techniques outside the scope of OSPRay and SWR. Knoll also observed that about half of the non-finalists – at least 50 percent – used OSPRay or CPU-based visualization in some fashion. “Before,” he said, “people knew that OSPRay existed – now they just use it by default in production.”  So, unlike 2015, CPU-based visualizations are no longer a contrary view.

An exascale requirement

The idea behind SDVis is that larger data sets imply higher resolution (and therefore quality) that is too big for typical GPU memory. Focusing directly on the needs of large scale visualization rather than first targeting gaming means that SDVis software components can be designed to utilize massive-memory hardware and algorithms that scale as needed across the nodes in a cluster or inside a computational cloud.

Massive data poses a problem as it simply becomes impractical from a runtime point of view to move it around or keep multiple copies. It just takes too much time and memory capacity. This makes in-situ visualization (which minimizes data movement by running the visualization and simulation software on the same hardware) a “must-have.”  As I like to say, “A picture is worth an Exabyte”.

Eliminating data movement with in-situ visualization is a hot topic in the scientific literature and is now viewed by experts as a technology requirement for visualization in the exascale era. The paper “An Image-based Approach to Extreme Scale In Situ Visualization and Analysis” by James Ahrens et al. quantifies the data movement challenge as follows: “Imagery is on the order of 10**6 in size, whereas extreme scale simulation data is on the order of 10**15 in size.” Nine orders of magnitude is significant.

Ahrens explained, “We believe very strongly that in-situ is a requirement for exascale supercomputing.” More specifically, “For exascale, we need to be portable across all platforms. It’s an IO and memory capacity issue.” Knoll agrees that in-situ visualization is a requirement, “the old way of business has to change.”

Managing success: CPU-based SDVis robustly encompasses new algorithmic and software approaches

Dr. Knoll points out that in-situ visualization encompasses a spectrum of technologies, not just software alone. He references the 3D XPoint and Intel Omni-Path architecture. Jointly developed by Micron and Intel, 3D XPoint is a non-volatile storage media that can be used as storage or to augment main memory as the media is byte-addressable. Intel Omni-Path is a high-bandwidth, low-latency communications architecture created by Intel to increase performance and decrease cost.

“Memory is key,” Knoll stresses. He points out that, “An Intel Xeon Phi processor can support up to 24x more DRAM than an equivalent single GPU (NVIDIA Tesla P100 with 16 GB RAM), and an Intel Xeon workstations (e.g., the Brickland-EX platform with 6 TB) up to 384x more. With 3D XPoint the cost of this ‘memory’ will decrease substantially, which goes hand in hand with the benefits of big data runtime algorithms where it does not cost substantially more to access (and render) 6 TB or data than 16 GB of data.”

Knoll envisions 3D XPoint working as an in-core file-system at scale that blurs the line between RDMA, in-situ visualization, and distributed file-systems. One example is the CORAL project that, “leverages Intel Crystal Ridge [now known as 3D XPoint] non-volatile memory technology that is configured in DDR4 compatible DIMM form factor with processor load/store access semantics on CORAL point design compute nodes. This software design will allow applications running on any CORAL point design compute node to have a global view of and global access to Crystal Ridge that is on other compute nodes.”

“This technology gets me very excited,” Knoll says, noting the importance of the communications fabrics in making fast distributed memory a reality.

Focusing visualization solutions on data size rather than gaming usage means that SDVis software components can be designed to utilize massive-memory hardware and scale as needed across the nodes in a cluster or inside a computational cloud. This frees developers to design for the user rather than the hardware.

Figure 2: The Los Alamos team won the visualization award at SC16 for their SDVis based work

The transition from OpenGL targeted hardware rasterization to CPU-based rendering means that algorithm designers can exploit large memory (100’s of GBs or larger) visualization nodes to create logarithmic runtime algorithms.

Dr. Knoll stresses the importance of logarithmic runtime algorithms (a subtle but key technical point) as users are faced with orders of magnitude increases in data sizes on the big supercomputers. Logarithmic runtime algorithms are important for big visualizations and exascale computing as the runtime increases slowly (e.g. logarithmically) even when data sizes increase by orders of magnitude. Such algorithms tend to consume large amounts of in-core memory to hold the data and associated data structures. Thus memory capacity and latency are two key hardware metrics.

Research at the University of Utah [PDF] shows a single large memory (3 terabyte) workstation can deliver competitive and even superior interactive rendering performance compared to a 128-node GPU cluster; this is paradigm-changing. The group is exploring in-situ visualization using P-k-d trees and other fast, in-core approaches [PDF]. This project at the University of Utah showed that large “direct” in-core techniques are not only viable, but are at the bleeding edge of visualization research.

Figure 3: Uintah combustion simulation visualized with VisIt using OSPRay

Our design efforts on OSPRay includes the recognition that our software cannot – and does not – exist in a vacuum. The challenge is to provide sufficient modularity so researchers can adapt the package without having to touch the golden build source code. In other words, OSPRay is designed so researchers can explore new approaches without breaking the code for everyone. Our solution was to extend OSPRay with the aptly named ‘modules’ capability, which first appeared in v1.2.0. In using modules, the University of Utah team notes that modules provide a logical pairing between algorithm and data where researchers can: (1) write a module and (2) pair it with distributed parallel data processing and rendering API such as the Argonne vl3 volume rendering library. Ultimately, this can allow simpler workflows and more efficient visualization of specific large problems, such as materials and cosmology data. By design, successful and widely-utilized modules can be evaluated by the OSPRay team across a number of platforms as possible additions to the main body of the OSPRay code. Such accessibility and portability across CPU platforms highlights the adaptable yet robust characteristics of SDVis software.

Education will likely increase the rate of adoption

The adoption rate over the past year has been phenomenal, but we expect it to increase even further. As Dr. Knoll stated, “2016 is the year OSPRay became used in practice and production.” As a production visualization tool for scientific computing, OSPRay and more generically CPU-based SDVis has clearly come of age. Integration into packages such as ParaView and VisIt has made CPU-based rendering mainstream, which in turn means that using a CPU for visualization can no longer be considered a contrary viewpoint; it’s becoming the norm.

It is expected that education will likely increase the rate of adoption. A number of excellent educational resources are available online. For example, view the 2016 Intel HPC Developer software visualization track videos to delve more deeply into the technology and third-party use cases. Of course, hands-on experience and interacting with peers is always of value. Such interactions can be had at the IXPUG May 2017 Visualization workshop at the Texas Advanced Computing Center. Immediate hands-on experience can also be had simply by working with VisIt and ParaView, or downloading the OSPRay code from github and the OpenSWR code via the Mesa3D website.  Further background and up to date information about Software Defined Visualization is available at our IDZ (Intel Developer Zone) SDVis landing page., and in Chapter 17 of my Morgan Kaufman published book Intel Xeon Phi High Performance Programming: Knights Landing Edition.

To utilize CPU-based SDVis in your software, look to the following packages: (1) the OSPRay scalable, and portable ray tracing engine; (2) the Embree library of high-performance ray-tracing kernels; and (3) OpenSWR, a drop-in OpenGL replacement, highly scalable, CPU-based software rasterizer all provide core functionality for current SDV applications.

About the Author

Jim Jeffers is a Principal Engineer and engineering leader at Intel who is passionate about world changing technology as well as author and industry expert on parallel computing hardware.

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!

Data Vortex Users Contemplate the Future of Supercomputing

October 19, 2017

Last month (Sept. 11-12), HPC networking company Data Vortex held its inaugural users group at Pacific Northwest National Laboratory (PNNL) bringing together about 30 participants from industry, government and academia t Read more…

By Tiffany Trader

AI Self-Training Goes Forward at Google DeepMind

October 19, 2017

DeepMind, Google’s AI research organization, announced today in a blog that AlphaGo Zero, the latest evolution of AlphaGo (the first computer program to defeat a Go world champion) trained itself within three days to play Go at a superhuman level (i.e., better than any human) – and to beat the old version of AlphaGo – without leveraging human expertise, data or training. Read more…

By Doug Black

Researchers Scale COSMO Climate Code to 4888 GPUs on Piz Daint

October 17, 2017

Effective global climate simulation, sorely needed to anticipate and cope with global warming, has long been computationally challenging. Two of the major obstacles are the needed resolution and prolonged time to compute Read more…

By John Russell

HPE Extreme Performance Solutions

Transforming Genomic Analytics with HPC-Accelerated Insights

Advancements in the field of genomics are revolutionizing our understanding of human biology, rapidly accelerating the discovery and treatment of genetic diseases, and dramatically improving human health. Read more…

Student Cluster Competition Coverage New Home

October 16, 2017

Hello computer sports fans! This is the first of many (many!) articles covering the world-wide phenomenon of Student Cluster Competitions. Finally, the Student Cluster Competition coverage has come to its natural home: H Read more…

By Dan Olds

Data Vortex Users Contemplate the Future of Supercomputing

October 19, 2017

Last month (Sept. 11-12), HPC networking company Data Vortex held its inaugural users group at Pacific Northwest National Laboratory (PNNL) bringing together ab Read more…

By Tiffany Trader

AI Self-Training Goes Forward at Google DeepMind

October 19, 2017

DeepMind, Google’s AI research organization, announced today in a blog that AlphaGo Zero, the latest evolution of AlphaGo (the first computer program to defeat a Go world champion) trained itself within three days to play Go at a superhuman level (i.e., better than any human) – and to beat the old version of AlphaGo – without leveraging human expertise, data or training. Read more…

By Doug Black

Student Cluster Competition Coverage New Home

October 16, 2017

Hello computer sports fans! This is the first of many (many!) articles covering the world-wide phenomenon of Student Cluster Competitions. Finally, the Student Read more…

By Dan Olds

Intel Delivers 17-Qubit Quantum Chip to European Research Partner

October 10, 2017

On Tuesday, Intel delivered a 17-qubit superconducting test chip to research partner QuTech, the quantum research institute of Delft University of Technology (TU Delft) in the Netherlands. The announcement marks a major milestone in the 10-year, $50-million collaborative relationship with TU Delft and TNO, the Dutch Organization for Applied Research, to accelerate advancements in quantum computing. Read more…

By Tiffany Trader

Fujitsu Tapped to Build 37-Petaflops ABCI System for AIST

October 10, 2017

Fujitsu announced today it will build the long-planned AI Bridging Cloud Infrastructure (ABCI) which is set to become the fastest supercomputer system in Japan Read more…

By John Russell

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

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

Intel Debuts Programmable Acceleration Card

October 5, 2017

With a view toward supporting complex, data-intensive applications, such as AI inference, video streaming analytics, database acceleration and genomics, Intel i Read more…

By Doug Black

Reinders: “AVX-512 May Be a Hidden Gem” in Intel Xeon Scalable Processors

June 29, 2017

Imagine if we could use vector processing on something other than just floating point problems.  Today, GPUs and CPUs work tirelessly to accelerate algorithms Read more…

By James Reinders

NERSC Scales Scientific Deep Learning to 15 Petaflops

August 28, 2017

A collaborative effort between Intel, NERSC and Stanford has delivered the first 15-petaflops deep learning software running on HPC platforms and is, according Read more…

By Rob Farber

How ‘Knights Mill’ Gets Its Deep Learning Flops

June 22, 2017

Intel, the subject of much speculation regarding the delayed, rewritten or potentially canceled “Aurora” contract (the Argonne Lab part of the CORAL “ Read more…

By Tiffany Trader

Oracle Layoffs Reportedly Hit SPARC and Solaris Hard

September 7, 2017

Oracle’s latest layoffs have many wondering if this is the end of the line for the SPARC processor and Solaris OS development. As reported by multiple sources Read more…

By John Russell

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

Google Releases Deeplearn.js to Further Democratize Machine Learning

August 17, 2017

Spreading the use of machine learning tools is one of the goals of Google’s PAIR (People + AI Research) initiative, which was introduced in early July. Last w 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

Graphcore Readies Launch of 16nm Colossus-IPU Chip

July 20, 2017

A second $30 million funding round for U.K. AI chip developer Graphcore sets up the company to go to market with its “intelligent processing unit” (IPU) in Read more…

By Tiffany Trader

Leading Solution Providers

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

Amazon Debuts New AMD-based GPU Instances for Graphics Acceleration

September 12, 2017

Last week Amazon Web Services (AWS) streaming service, AppStream 2.0, introduced a new GPU instance called Graphics Design intended to accelerate graphics. The Read more…

By John Russell

EU Funds 20 Million Euro ARM+FPGA Exascale Project

September 7, 2017

At the Barcelona Supercomputer Centre on Wednesday (Sept. 6), 16 partners gathered to launch the EuroEXA project, which invests €20 million over three-and-a-half years into exascale-focused research and development. Led by the Horizon 2020 program, EuroEXA picks up the banner of a triad of partner projects — ExaNeSt, EcoScale and ExaNoDe — building on their work... Read more…

By Tiffany Trader

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

Cray Moves to Acquire the Seagate ClusterStor Line

July 28, 2017

This week Cray announced that it is picking up Seagate's ClusterStor HPC storage array business for an undisclosed sum. "In short we're effectively transitioning the bulk of the ClusterStor product line to Cray," said CEO Peter Ungaro. Read more…

By Tiffany Trader

Intel Launches Software Tools to Ease FPGA Programming

September 5, 2017

Field Programmable Gate Arrays (FPGAs) have a reputation for being difficult to program, requiring expertise in specialty languages, like Verilog or VHDL. Easin Read more…

By Tiffany Trader

IBM Advances Web-based Quantum Programming

September 5, 2017

IBM Research is pairing its Jupyter-based Data Science Experience notebook environment with its cloud-based quantum computer, IBM Q, in hopes of encouraging a new class of entrepreneurial user to solve intractable problems that even exceed the capabilities of the best AI systems. Read more…

By Alex Woodie

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