Intel EVP confesses dreams of an HPC future

In an interview with El Reg, Intel EVP Sean Maloney — a man many think is the alpha dog in line for the CEO slot once Otellini hits the bricks — comes clean about his vision for an HPC future

Intel logo“So, I thought about where processing power will go. If we can deliver 10,000 times the processing power we have today, you would be able to predict weather patterns. What if I knew what the wind speed will be when I go rowing? What if you knew in Bombay when it will rain? Can you imagine what you could do with water savings? It would be so cool.”

…”We are planning and investing for that type of future,” Maloney said. “When we talk about things like predicting when it rains, we want that to run on Intel machines.”

For the top-end server stuff, Intel has no fears around how customers will use the multi-core chips, er, flooding the market.

“In the high-end segment, there is an infinite requirement for more compute power,” Maloney said. “The high order bit on that conversation is can you do it without using a nuclear power station.”

Sometimes I dream about food, just in case Ashlee wants to do a story on me.

Berkeley researchers propose 200 PFLOPS super from embedded processors

The Australian iTnews (yes, the capitalization is both correct and odd) was the most interesting pointer I found to this story.

Lawrence Berkeley Lab researchers shooting for a practical solution for modeling climate at a resolution of 1 km came up with a system design based on low-power embedded processors. You know, the kinds of processors you find in cell phones and other consumer electronics.

They estimate that building a conventional supercomputer capable of modelling [sic] clouds at a 1km scale would cost about $1bn…[and] require 200 megawatts of electricity to operate, enough to power a small city of 100,000 residents.

In a paper entitled ‘Towards Ultra-High Resolution models of Climate and Weather’ the researchers present an alternative that would cost less to build and require less electricity to operate.

They conclude that a supercomputer using about 20 million embedded microprocessors would deliver the results and cost US$75 million to construct.

This “climate computer” would consume less than 4-megawatts of power and achieve a peak performance of 200 petaflops.

Go, boffins, go. Full article here.

Mercury puts 100 GFLOPS in your hand

Mercury Computer Systems announced this week that they’ve released a new compute platform that puts 100 GFLOPS in a package you can hold in your hand weighing “less than 10 pounds.”

The PowerBlock(TM) 50 system from Mercury Computer Systems is a high-performance, ultra-compact embedded computer designed for maximum performance in a minimal footprint, for small platforms in the 6- to 10-pound range. Optimized for real-time image, sensor, and signal processing and ruggedized for harsh environments, it is fully integrated and programmable, with state-of-the-art liquid cooling. The PowerBlock 50 extends Mercury’s robust offering of computing solutions available in very small to large form factors to suit a broad range of customer applications.

Now, let’s link a bunch of these together and run a Top 10 system in an abandoned hangar in the desert for the ultimate in green HPC.

Platform announces FS unit

Platform Computing announced yesterday that they’ve formed a new business unit

Platform logoTo build on the momentum and capitalize on opportunities in Financial Services, the company created a unit dedicated to this specialized market sector. With consistent strong growth across all major geographies including North America, Europe and Asia, Financial Services continues to be a global driver for Platform. Platform’s Financial Services Business Unit will be based in New York City.

SGI announces Q3 results; losses widen

SGI announced its Q3 financial results yesterday, tag teaming it with the 20,480 NASA system announcement and then bracing it with the PFLOPS deal announced today.

Evidently they were trying to soften the news that the company’s revenue dropped $11M from Q2, with losses widening from $30.8M in Q3 to $40.6M this quarter

One bright(er) spot was the growth of orders with grew 50% year over year, and backlogs

Backlog at the end of the third quarter of 2008 grew to $133.9 million compared to $95.8 million at the end of the second quarter of fiscal 2008, the highest backlog level in the past five quarters.

SGI stock is down over 10% in mid-afternoon trading today.

NASA and SGI kick it up a notch: 1 PFLOPS in the Pleiades

Following closely on the heels of yesterday’s announced 20,480 core SGI ICE system headed to NASA, today we have the announcement of a 1 PFLOPS system named Pleiades (info on _the_ Pleiades here).

SGI logoUnder the terms of a Space Act Agreement, NASA will work closely with Intel and SGI to increase computational capabilities for modeling and simulation at the NASA Advanced Supercomputing (NAS) facility at NASA’s Ames Research Center, Moffett Field, Calif.

…NASA Ames, Intel and SGI will work together on a project called Pleiades to develop a computational system with a capacity of one Petaflops peak performance (1,000 trillion operations per second) by 2009 and a system with a peak performance of 10 Petaflops (10,000 trillion operations per second) by 2012.

Top500 Rumor Mill

top500Those that track the Top500 list are most certainly awaiting the latest summer release from Germany. Given that there have been several large announcements surrounding big compute in the previous six months, we decided to throw together a series of rumor propositions regarding some possible shakeups in the top ten.

The most obvious contender for the top 10 is the Ranger system housed with our friends at TACC. Regardless of how one feels about AMD, quad core methodologies, Sun Microsystems or Infiniband; you simply can’t ignore 62,000 scalar cores. If they can eek out 70% of peak performance, they should do quite well on the list. [Keep in mind, Goto spends his days here].

Argonne has taken ownership of a new BlueGene/P. Again, regardless on what your opinion is of the IBM BlueGene architecture, they perform well on Linpack. They still hold four of the top ten spots.

We’ve mentioned Sun and IBM, but what about the Purple People Eater, SGI? More specifically, what happened to Columbia? For residents of Mars, Columbia is the fabled system located at the NASA Advanced Supercomputing Center in Mountain View, California. This Itanium behemoth broke 10,000 cores before it was fashionable to do so. Word is, they’ve completed an initial upgrade to Columbia with a few Altix 4700. To complement the upgrade, they’ve rerun Linpack across the entire cluster. This could prove to be fairly impressive.

The most interesting of rumors we’ve had filling our inbox is in regard to Los Alamos National Lab’s RoadRunner hybrid system. For those n00b readers out there, RoadRunner is working on its third phase of integrating AMD Opteron and IBM Cell processors. LANL and IBM had some pain with initially integrating an untested platform, but insiders say they’ve made it over the initial hurdles. Word on the grapevine is, they’re poised to take the number one spot away from their LLNL brethren.

Not specifically mentioned [mostly due to lack of information] are systems in China, Oak Ridge, DoD MSRCs and GFDL. All of these locations have taken delivery of large systems since the previous Top500 deadline. No word on how well these integrations are proceeding.

Regardless of your leanings regarding compute architectures, keep a keen eye this June when the new list is released.

SGI Win’s NASA’s Next Big Toy

Alongside SGI’s earnings press release delivered this afternoon, they also announced a large supercomputing win with NASA. There aren’t a great deal of details related to the deal, but this is rumored to be the NASA Advanced Supercomputing Columbia replacement.

They did, however, release some rough system specs. The system will be an Altix ICE platform with 20,480 cores and 20TB of memory. Go go gadget cluster!

[Ed: the system will also apparently be watercooled and, as Ashley Vance of The Register predictably notes, marks a departure from NASA from SGI’s Itanic…er, Itanium…based systems.]

University of Houston Builds 8 Megapixel Viz Theater

HPCWire: The Texas Learning & Computation Center [TLC2] located at the University of Houston has announced a contract with Mechdyne Corporation to build a 34-seat, 8 megapixel stereographic theater.  The theater, powered by two Sony SXRD projectors, will be available to all academic departments at the University of Houston as well as all UH/TLC2 partners.

While our first generation visualization laboratory was set up as a working space for individuals or small groups, the new theater is designed to provide unique capability for both researchers and developers of rich graphics imagery to share data and images,” said Erick Engquist, manager of the Visualization Laboratory. “In addition to the traditional fields that utilize computer visualization, such as medical imaging, molecular dynamics and geophysical sciences, the theater will enable faculty and students in our schools of architecture and fine arts to share their work in new and exciting ways.”

Read the full article at HPCWire.

Homegrown HPC

linuxworldLinuxWorld Australia has an interesting article covering homegrown high performance computing.  Many of us have been down this road, however painful.  They go as far as to provide four realistic and successful implementations in production today.

The four case studies assembled here illustrate the pain and complexity of building a successful HPC environment, including the sensitive hardware and software dependencies that affect performance and reliability, as well as the painstaking work that goes into parallelizing serial apps to work successfully in a clustered environment.

I won’t go into all the gory details, so read the full article here.

Verari Hits Milestone Blade Shipment

verariVerari has announced that they completed a milestone shipment of its blade-based computing solutions.  An undisclosed customer took delivery of over 500 of Verari’s SB5550XL blade servers as a platform for enterprise and web 2.0 workloads.

Having a worldwide technology leader successfully complete a major installation of our power-efficient blade-based storage and server solutions validates our product and quality leadership,” said David B. Wright, CEO and Chairman of the Board. “This milestone is also indicative of the importance of a hybrid design that can deliver both storage density and high-performance processing while providing up to 50% savings in energy use and up to 77% better floor space utilization then traditional storage and server systems.”

For more info, read the full post here.

Larrabee engineer: it’s all about the rasterization

Last week ars technica posted a piece about Larrabee engineer Tom Forsyth’s personal blogging on Larrabee’s biggest strength: rasterization

Larrabee, a massively multicore x86 chip due out in 2010 as a discrete graphics card, has sparked a lot of interest from tech journalists, along with a great deal of confusion. The chip has had a great deal of press as a potential ray-tracing chip, physics cruncher, or HPC chip, but Forsyth emphasized the 2010 release and primary design goals were focused on rasterization because it’s the only way to “render the huge range of DirectX and OpenGL games out there.”

HPC enables discovery of new blood pressure drug

Also at the new HPCwire today, news that U of Florida researchers used HPC to find a new drug that lowers blood pressure and prevents heart and kidney damage, at least in rats. More research in humans coming.

Researchers used one of the world’s most powerful supercomputers to process 140,000 prospective drug compounds in a matter of weeks. The computer predicted which molecules would be most likely to enhance the activity of ACE2, rotating them in thousands of different orientations to see how they would bind to certain pockets on the enzyme’s surface.

…After hitting on the “lead” compound, UF researchers then tested it in hypertensive rats that had developed fibrosis of the heart and kidney. The animals received the drug for two weeks. Tissue samples from treated animals revealed a significant decrease in fibrosis of the heart, kidney and blood vessels, said Ostrov, who described the findings as “striking and reproducible.”

And the good news on this one keeps coming, apparently

…Early results also show the compound inhibits inflammation, which has significant implications for a number of human diseases, including autoimmune diseases such as type 1 diabetes and rheumatoid arthritis as well as other diseases involving fibrosis, such as Alzheimer’s, Ostrov said.

As the old SNL sketch goes, “it’s a non-dairy whipped topping AND a floor wax.” Anyway, good stuff.

NCSA parallel workshop

Posted at HPCwire today

The National Center for Supercomputing Applications (NCSA / http://www.ncsa.uiuc.edu) and the Institute for Advanced Computing Applications and Technologies (IACAT / www.iacat.uiuc.edu) will present a Workshop on Programming Massively Parallel Processors (PMPP) on July 10, 2008, in Urbana-Champaign, Ill.

…The workshop will bring together researchers, industry, and users concerned with the issue of programming multi-core and many-core architectures for productive use in applications ranging from desktop to high-performance computing systems. The workshop will include presentations from industry leaders and early technology adopters and will stimulate the dialogue among researchers about programming models, languages, and tools that the community is lacking in order to take advantage of the emerging multi-core and many-core architectures.

Sun reports 3rd quarter results, posts loss

Lots of ink spilled in the IT press last week over Sun’s quarterly results. Revenues for the quarter that ended in March were down $17M and came in $100M under analysts expectations, according to coverage at FT.com. Sun posted a loss of $34M after taxes, or 4 cents a share. The 3Q2007 result was a net income of 7 cents a share.

The real problem with all of this is that the Sun leadership team has been going on about 10% year over year growth. From coverage at TheStreet.com

Sun’s weak quarterly results marked the first major failure under the leadership of Jonathan Schwartz and Mike Lehman, who took over as CEO and CFO, respectively, in 2006.

The pair returned Sun to profitability last year, after several years of losses, and promised investors that revenue growth and a 10% operating margin were both around the corner.

But the message on Thursday was that neither goals appeared in sight anymore.

Sun projected that revenue in the fourth quarter would be flat year-over-year. And instead of the 10% operating income targeted for next fiscal year, Sun is now aiming for 7%.

Sun is planning to offset the losses with a layoff of from 1,500 to 2,500 employees (of about 34,000 total).