Roadrunner and Jaguar, the DOE supercomputers that launched the petaflop era last year, will soon be eclipsed by new machines more than ten times as powerful. IBM and the US National Nuclear Security Administration (NNSA) announced on Tuesday that in 2011 Lawrence Livermore National Laboratory will install a 20 petaflop system to provide computational support for the country’s aging nuclear weapons.

Building on its Blue Gene heritage, IBM will deliver “Dawn,” a 500 teraflop Blue Gene/P system in the first quarter of this year, followed by “Sequoia,” a 20 petaflop next-generation Blue Gene/Q machine for 2011. Sequoia is expected to officially go online in 2012. The new machines will take over Lawrence Livermore’s weapon simulation codes that are being maintained under the Advanced Simulation and Computing (ASC) Program. Currently this work is being done with the existing capability supercomputers at the lab: the 100 teraflop ASC Purple and the 600 teraflop Blue Gene/L.

Dawn will act as an interim platform for porting and scaling the weapons codes. Once the Blue Gene/Q super comes online, those codes will be moved over to the bigger machine for production. The Dawn machine is in the process of being built right now, with about half of the machine already wired together at Lawrence Livermore. The lab is planning on getting the rest of the hardware over the next few months, with system acceptance scheduled for April.

Using Dawn as a stepping stone to Sequoia is possible since, unlike Blue Gene/L, both Blue Gene/P and Blue Gene/Q support node-level cache coherency, which allows for SMP-style programming. Especially for the weapons code, mapping one MPI task per core would be a real challenge, but going to a mixed SMP-message passing model — shared-memory parallelism within the nodes and distributed parallelism across the nodes — is much more practical.

Not only will Sequoia be more than ten times as powerful as the current crop of petaflop supercomputers, its energy efficiency will be much improved. According to IBM Deep Computing VP Dave Turek, Sequoia will consume around 6 megawatts, yielding an energy efficiency ratio of over 3,000 MFLOPS/watt*. That represents a 7X improvement over the Blue Gene/P generation (440 MFLOPS/watt*), and is even better than the Cell-based Roadrunner system at Los Alamos (587 MFLOPS/watt*). For a starker comparison, the 1.6 petaflop Opteron-based Jaguar supercomputer installed at Oak Ridge National Laboratory uses about 8.5 megawatts (188 MFLOPS/watt*).

When Sequoia arrives in the first half of 2011, space is going to be at a premium in the lab’s Terascale Simulation Facility, (which already houses ASC Purple and the Blue Gene/L system) but power is going to be the real problem. Although both new Blue Genes are much more energy efficient than their predecessors, the lab is planning to more than double the facility’s power — from 12.5 to 30 megawatts.

IBM is not releasing low-level details of the Blue Gene/Q architecture. However, since Sequoia will be composed of 98,304 compute nodes and contain a total of 1.6 million cores, one can surmise that a Blue Gene/Q node will contain 16 cores. Whether this is implemented as one 16-core chip or two 8-core chips (or even four quad-core chips) remains to be seen. Since Sequoia will sport 1.6 petabytes of memory, each node stands to have 16 GB. The current Blue Gene/P technology offers 4 cores and 4 GB of main memory per node.

At 20 petaflops, Sequoia will be 160 times as powerful as Lawrence Livermore’s ASC Purple and 17 times as powerful as its current Blue Gene/L, giving scientists a lot more computing cycles for weapons simulations and basic science research. “It’s been an interesting journey,” notes Turek. “When you think back to when the ASCI [now ASC] program was launched in the 90s and what the aspirations were for FLOPs back then versus where we are today, I think we’ve exceeded everyone’s expectations.”

Indeed. Considering the original supercomputers under the ASC program (i.e., ASCI Blue Pacific at 3.9 teraflops and ASCI White at 12.3 teraflops) don’t even show up on today’s TOP500 list, the new systems represent a completely different class of capability for the stockpile stewardship program. Mark Seager, who manages the Platforms Program for the ASC Program at Lawrence Livermore and led the team that wrote the RFP for the new machines, says Sequoia will enable a new level of predictive science.

Toward that end, the lab will be enhancing the existing weapons codes with “uncertainty quantification” (UQ) methods. Seager says this is a relatively new branch of science that allows researchers to apply a lot of physics parameters to the simulations. With this model, researchers will be able to quantify the errors associated with simulation results. Once the largest sources of errors are known, the models can be systematically refined to enhance the predictive capabilities. Unfortunately, UQ is computationally expensive, so only limited numbers of simulations can be attempted on existing hardware.

“On [ASC] Purple we were able to do a UQ study on one weapons system in about a month with approximately 4,400 calculations, some of which took up the maximum practical size of the machine, which is 8,192 MPI tasks,” explains Seager. “With Sequoia, multiply that capability by somewhere between 12 and 24X.”

But MPI applications tend to be very sensitive to hardware or software failures, so completing a fault-free run is going to be challenging at the scale of a million-plus cores. To address the resiliency issue, Seager says they’ll be applying “ensemble” calculations to their codes. In the ensemble method, the same algorithm can be run thousands of time with different sets of parameters. Using this approach, isolated failures on a small number of calculations can be tolerated without sacrificing the integrity of the whole application. It’s analogous to the way many Web applications like search engines operate today.

Sequoia’s second mission will be to support basic science at scale, where scientists are looking to achieve 20 to 50 times the capability that is provided by the existing Blue Gene/L system. Along with the extra capability Sequoia will provide the weapons codes migrating from ASC Purple, Lawrence Livermore stands to leapfrog rather decisively into the petascale era. Says Seager: “It is probably the single largest jump in computing power that the lab has ever seen.”

*The original version of this article incorrectly expressed the energy efficiency ratios at FLOPS/watt, instead of MFLOPS/watt.