Since 1986 - Covering the Fastest Computers
in the World and the People Who Run Them
Visit additional Tabor Communication Publications
Blog: From the Editor
From the Editor | Main Blog Index
October 28, 2010
New Chinese GPGPU Super Outruns Jaguar
The end of the US dominance at the top of the TOP500 appears to be at hand. Tianhe-1A, a new Chinese supercomputer powered by over 7,000 NVIDIA Tesla GPUs has recorded a Linpack score of 2.507 petaflops. That would beat out Oak Ridge National Lab's 1.759 petaflop Jaguar machine, the current TOP500 title holder, by a wide margin.
It would also be the first time a non-US supercomputer held the number one spot in six years. From June 2002 to June 2004, Japan's Earth Simulator was the fastest supercomputer in the world. In September 2004, it yielded its title to the new kid on the block -- IBM's Blue Gene/L. The US has never looked back.
Until now.
The concentrated performance available in high end discrete GPUs has opened up the petaflop club for a lot more players. But to get to the top of the heap, you need thousands of GPUs. Tianhe-1A, for example, sports 7,168 of them, in this case, NVIDIA Tesla M2050 (Fermi) GPUs. These represent the lion's share of FLOPS in the system, despite the presence of accompanying 14,336 CPUs. Tianhe-1A also comes with 262 TB of memory and 2 PB of Lustre-based storage.
Although the Linpack performance is a stunning 2.5 petaflops, the system left a lot of potential FLOPS in the machine. Its peak performance is 4.7 petaflops, yielding a Linpack efficiency of just over 50 percent. To date, this is a rather typical Linpack yield for GPGPU-accelerated supers. Because the GPUs are stuck on the relatively slow PCIe bus, the overhead of sending calculations to the graphics processors chews up quite a few cycles on both the CPUs and GPUs.
By contrast, the CPU-only Jaguar has a Linpack/peak efficiency of 75 percent. Even so, Tianhe-1A draws just 4 megawatts of power, while Jaguar uses nearly 7 megawatts and yields 30 percent less Linpack.
Despite the exotic nature and stature of Tianhe-1A system, it is targeted for typical high performance computing applications including oil exploration, equipment development, biomedical research, animation design, weather forecasting, financial risk analysis, remote sensing, materials research, and the like.
Like its immediate ancestor, Tianhe-1, the Tianhe-1A system was developed by the National University of Defense Technology (NUDT) and is being housed at National Supercomputer Center in Tianjin. Tianhe-1, though, was built using AMD GPUs, specifically 2,560 dual-GPU ATI Radeon HD 4870 X2 processors. That system was launched in 2009. Somewhere along the line, NUDT decided to switch horses and go with NVIDIA gear. Since there is currently no CUDA port for AMD GPUs, software compatibility between the two systems is going to problematic, unless they go the OpenCL route.
China's enthusiasm for GPGPUs has propelled its supercomputing capacity significantly. Even before Tianhe-1A, the country claimed three systems in the top 20: Nebulae at number 2, Tianhe-1 at number 7, and Mole 8.5 at number 19, all of which use GPUs. The US, Germany, and the UK currently have no GPU-equipped systems on the TOP500.
Although the official list results won't be revealed until the middle of November, it's doubtful if a secret supercomputer is lying in wait ready to challenge a 2.5 petaflop machine. But we'll find out soon enough.
Posted by Michael Feldman - October 28, 2010 @ 12:40 AM, Pacific Daylight Time
Discussion
There is 1 discussion item posted.
Tianhe-1A/Jaguar FLOPS/W
Submitted by koeningerk
on Oct 28, 2010 @ 11:37 AM EDT
TCO? Michael notes Jaguar has better sustained/peak Linpack efficiency compared to Tianhe-1A (75 and 50% respectively). He hinted that Jaguar is also more efficient in MFLOPS/W but did show the numbers: 357 vs. 450 MFLOPS/W (Jaguar better). My studies on Fermi/x86 price/performance show Fermi clusters are likely to deliver twice the Linpack FLOPS/$ (and higher ratios with with more GPUs/CPU). Lacking an actual price reported for Tianhe-1A, let's assume 2x P/P for Tianhe-1A. So which is more important, saving 20% on the energy bill or saving half on the capital bill? At $0.10 per KWH, 7 MW costs $700 per hour or ~$6 M per year. So Jaguar would save about $1.25 M per year in operational costs but would cost over $100 M more in capital costs. Clearly Tianhe-1A wins on total cost of operation (TCO), when running Linpack
I'm curious how much of the cost of Tianhe-1A is for NVIDIA and how much is for the x86 servers. When applications are not using the NVIDIA chips or use them poorly the P/P of Jaguar will be much better. So what are the TFLOPS rates on real applications and what percentage of Tianhe-1A's time will be spend on NVIDIA-efficient codes that matter? Those factor into a realistic measure of real TCO. For weather, seismic (oil), and biotech codes Tianhe-1A should win. For chemistry, fluid dynamics, and general HPC codes the picture is still murky.
Note, Roadrunner was the first to break the PFLOPS barrier, but its Linpack P/P was 3 times worse than x86!
Post #1
|
Join the Discussion |
Sponsored Links
One of the first HPC providers installing a 4X NVIDIA Kepler K-20 cluster. Invites you to a free evaluation on Seneca’s NVIDIA K20 Kepler cluster, pre-loaded with AMBER, NAMD, LAMMPS
Webinar: Programming Heterogeneous X64+GPU Systems Using OpenACC
Join Michael Wolfe as he compares the advantages and costs of using both low-level models and the directive-based OpenACC model for programming accelerated heterogeneous systems. Registration is free.
High-Performance Computing in Action
Businesses that want to be on the cutting edge of their industries are increasingly turning to high-performance computing (HPC) solutions to handle complex compute processes and speed up their rate of innovation. Download this Executive Brief to see how businesses in energy, life sciences and entertainment put HPC solutions to work in their operations.
- Memory Debugging Challenges in Parallel and Distributed Applications
- Receive Higher Application Performance with end-to-end 56GbE
- Acer and Altair simplify cluster computing
- ClusterStor™ 1500 – High Performance Efficiency
- Visualize Historical Usage for Optimized Returns on HPC Investments
- PGI 2013: OpenMP 1.5X faster than GCC, new GNU-compatible OpenACC C++ compiler, CUDA 5 and Tesla K20 support, and more.
- White Paper: Building a Data-Intensive Cluster Supercomputer
- The Accelerated Genomics Cloud $25,000 Giveaway
RSS Feeds
Feeds by Topic
- Applications
- Developer Tools
- Interconnects
- Middleware
- Networks
- Processors
- Storage
- Systems
- Visualization
Feeds by Industry
Feeds by Content Type
- Features
- Blogs
- Short Takes
- Off the Wire
- Whitepapers
- Multimedia
- HPC Soundbite Podcasts
- HPCwire Job Bank
Subscribe to All Content
Michael Feldman
Michael Feldman is the editor of HPCwire.
Recent Comments
No Recent Blog Comments
Feature Articles
Exascale Advocates Stand on Nuclear Stockpiles
In quieter times, sounding the bell of funding big science with big systems tends to resonate further than when ears are already burning with sour economic and national security news. For exascale's future, however, the time could be ripe to instill some sense of urgency....
Read more...
NSF Forges Further Beyond FLOPs
In a recent solicitation, the NSF laid out needs for furthering its scientific and engineering infrastructure with new tools to go beyond top performance, Having already delivered systems like Stampede and Blue Waters, they're turning an eye to solving data-intensive challenges. We spoke with the agency's Irene Qualters and Barry Schneider about..
Read more...
CERN, Google Drive Future of Global Science Initiatives
Large-scale, worldwide scientific initiatives rely on some cloud-based system to both coordinate efforts and manage computational efforts at peak times that cannot be contained within the combined in-house HPC resources. Last week at Google I/O, Brookhaven National Lab’s Sergey Panitkin discussed the role of the Google Compute Engine in providing computational support to ATLAS, a detector of high-energy particles at the Large Hadron Collider (LHC).
Read more...
Short Takes
NASA Builds 'Climate in a Box'
May 23, 2013 |
The study of climate change is one of those scientific problems where it is almost essential to model the entire Earth to attain accurate results and make worthwhile predictions. In an attempt to make climate science more accessible to smaller research facilities, NASA introduced what they call ‘Climate in a Box,’ a system they note acts as a desktop supercomputer.
Read more...
Building Supercomputers with Raspberries
May 22, 2013 |
At some point in the not-too-distant future, building powerful, miniature computing systems will be considered a hobby for high schoolers, just as robotics or even Lego-building are today. That could be made possible through recent advancements made with the Raspberry Pi computers.
Read more...
Running Computational Fluid Dynamics in the Cloud
May 16, 2013 |
When it comes to cloud, long distances mean unacceptably high latencies. Researchers from the University of Bonn in Germany examined those latency issues of doing CFD modeling in the cloud by utilizing a common CFD and its utilization in HPC instance types including both CPU and GPU cores of Amazon EC2.
Read more...
Computing the Physics of Bubbles
May 15, 2013 |
Supercomputers at the Department of Energy’s National Energy Research Scientific Computing Center (NERSC) have worked on important computational problems such as collapse of the atomic state, the optimization of chemical catalysts, and now modeling popping bubbles.
Read more...
Sponsored Whitepapers
Best Practices in Big Data Storage
05/10/2013 | Cleversafe, Cray, DDN, NetApp, & Panasas | From Wall Street to Hollywood, drug discovery to homeland security, companies and organizations of all sizes and stripes are coming face to face with the challenges – and opportunities – afforded by Big Data. Before anyone can utilize these extraordinary data repositories, however, they must first harness and manage their data stores, and do so utilizing technologies that underscore affordability, security, and scalability.
Progress in Parallel: the Bull Parallel Programming Center
04/15/2013 | Bull | “50% of HPC users say their largest jobs scale to 120 cores or less.” How about yours? Are your codes ready to take advantage of today’s and tomorrow’s ultra-parallel HPC systems? Download this White Paper by Analysts Intersect360 Research to see what Bull and Intel’s Center for Excellence in Parallel Programming can do for your codes.
Sponsored Multimedia
SGI DMF ZeroWatt Disk Solution
In this demonstration of SGI DMF ZeroWatt disk solution, Dr. Eng Lim Goh, SGI CTO, discusses a function of SGI DMF software to reduce costs and power consumption in an exascale (Big Data) storage datacenter.
Cray CS300-AC Cluster Supercomputer Air Cooling Technology Video
The Cray CS300-AC cluster supercomputer offers energy efficient, air-cooled design based on modular, industry-standard platforms featuring the latest processor and network technologies and a wide range of datacenter cooling requirements.
Blogs by Topics
Blogs by Author
- James Reinders
- Gary Johnson
- Andrew Jones
- Tiffany Trader
- Addison Snell
- Nicole Hemsoth
- Charlie Catlett
- Sue Korn
- Michael Feldman
- Chris Willard
- Isaac Lopez
- Tom Tabor
- Cathy Davidson
- William Lu
- Michael Wolfe
- Dan Gatti
- Caroline Connor
HPC Blogroll
- Computing Research Policy Blog
- Interconnects
- Reed's Ruminations
- Scalability.org
- StorageMojo
- Sun HPC Watercooler
Featured Events
- June 13, 2013
- June 14, 2013
Think Beyond Plastics Conference
Berkeley, CA
United States
- June 16, 2013
- June 20, 2013
ISC'13
Leipzig,
Germany
.gif)
- June 17, 2013
- June 18, 2013
Forecast 2013
San Francisco, CA
United States
- June 20, 2013
- June 21, 2013
CIO North America
Marina Del Rey, CA
United States
- June 24, 2013
- June 25, 2013
The Trading Show Chicago
Chicago, IL
United States
- June 25, 2013
- June 26, 2013
TERATEC 2013 Forum
Paris,
France
