Supercomputers take efficiency up another notch
RISC cores and hybrids deliver most flops per watt
SC10 More than the hand-wringing over parallel computing, the mounting electricity bill is the limiting factor holding back the growth of petascale systems. The recurring joke at the SC10 supercomputing conference last week was that we cannot build exascale systems that require their own nuclear power plant to juice them up. And that means supercomputer system designers have to figure out how to do more flopping with fewer electrons.
The Green500 ranking of supercomputers is a twist on the Top 500 ranking, but instead of ranking the machines by their sustained performance on the Linpack Fortran matrix math benchmark, the machines are arranged by their energy efficiency (in terms of megaflops per watt) as they run that benchmark test. The November 2010 Green500 list, which you can see here , is not just a re-sort of the Top 500 list. Some machines that are very energy efficient are nonetheless small in terms of their teraflops and thus do not make the Top 500 ranking. Similarly, some relatively large and famous supers are so awful in terms of how much juice they use that they don't make the Green500 list. But generally speaking, there is a fair amount of overlap.
The Top 500 list is compiled by Erich Strohmaier and Horst Simon of the Lawrence Berkeley National Laboratory, Jack Dongarra of the University of Tennessee, and Hans Meuer of the University of Manheim. We told you all about the latest Top 500 last week . The Green500 list, created by Wu-chun Feng and Kirk Cameron of Virginia Tech, has only been published eight times (two or three times a year, generally) since November 2007; the Top 500 list comes out twice a year, but has been published for the past 18 years.
The greenest super on the November 2010 Green500 rankings is the prototype of the BlueGene/Q that Big Blue is building for Lawrence Livermore National Laboratory for installation in 2012. (See the drilldown  into the BlueGene/Q from El Reg for details on this super's design.)
The half-rack of BlueGene/Q super, which is currently running at IBM's Watson Research Center in New York, ranked 115th on the Top 500 list at 65.35 teraflops (with 8,192 Power cores running 1.6 GHz) and burned 38.8 kilowatts of juice. That works out to 1,684 megaflops per watt, giving the BlueGene/Q a considerable lead in terms of energy efficiency. You can see why LLNL is planning on ramping this machine up to 20 petaflops in 2012. If the efficiency holds, that 20 petaflops box with nearly 1.6 million cores will burn 7.4 megawatts of juice.
By way of comparison, the top-ranked Tianhe-1A super located at the National Supercomputing Center in Tianjin, China, has a sustained performance of 2.57 petaflops and burns just over 4 teraflops of juice with its hybrid Intel Xeon-Nvidia GPU architecture, for a 635.2 megaflops per watt ranking on the Green500 list. (That's the 11th most power-efficient super on the list).
Here's another interesting comparison. The Jaguar Cray XT5 super at the US Department of Energy's Oak Ridge National Laboratory, which is sorely in need of an upgrade, burns just under 7 megawatts to deliver its 1.76 petaflops using 224,162 Opteron cores, which works out to only 253.1 megaflops per watt but only an 81st ranking on the Green500 list. Cray XT5 customers can at double their power efficiency by moving to twelve-core Opteron 6100 processors, and maybe more if they move to XE6 frames with the faster "Gemini" XE interconnect. (It depends on the workload.) Even moving to sixteen-core Opterons in 2011 will only boost performance by around 33 per cent and probably flops per watt by around the same amount, which will only get a future XE6 machine to somewhere around 675 megaflops per watt. (That is assuming a constant clock speed for the Opteron processors and increasing core counts in the processor sockets.) You can see why Cray is trying to figure out how to interface GPU co-processors into its system design. Increasing x64 core counts are not going to do the energy efficiency trick.
The Tsubame 2 ceepie-geepie cluster at the Tokyo Institute of Technology in Japan was the third most powerful machine on the Top 500 list, at 1.19 petaflops on the Linpack test; that machine lashes three Nvidia M2050 GPU co-processors to every server node, and the GPUs are doing approximately 70 per cent of the Linpack calculations.
Feng and Cameron estimate that the Tsubame 2 cluster burns 1.24 megawatts, and when you do the math, it delivers 958.4 megaflops per watt. Even if you could get a ceepie-geepie cluster to offer the 80-ish percent efficiency of a traditional CPU cluster using a fast interconnect, the BlueGene/Q would still beat out Tianhe-1A and Tsubame 2 in terms of flops per watt. And that BlueGene/Q machine has similarly awful efficiency, with about half its flops going up the chimney right now, so there is plenty of room for IBM to do some tuning to get even better efficiency with BlueGene/Q before and after it is delivered to LLNL a little more than a year from now.
GPU co-processors are going to have to have a dramatic leap in performance in the same thermal envelope and the overhead of talking across the PCI-Express bus is going to have to be lowered for GPUs to compete. And the word from SC10 from both Nvidia and Advanced Micro Devices is that they are going to give massively cored machines like BlueGene/Q a serious run for their supercomputing money.
The third most energy-efficient super on the Green500 list doesn't have a lot of flops, but the EcoG super installed at the National Center for Supercomputing Applications does downshift to dual-core Core i3 processors running at 2.93 GHz to manage the Nvidia C2050 GPU co-processors in the cluster, eliminating lots of heat from the CPU side. The machine was rated at 33.6 teraflops, but burned 36 kilowatts of juice, for a power rating of 933.1 megaflops per watt.
Another prototype machine, the Fujitsu-designed K super going into Japan's Riken Advanced Institute for Computational Science, ranked fourth on the Green500. Formerly known as Project Keisoku and originally intended to scale up to 10 petaflops using a mix of vector processors made by NEC and Hitachi and scalar processors made by Fujitsu, the entire machine is now being built  only using Fujitsu's eight-core "Venus" Sparc64-VIIIfx processors and a 6D interconnect called Tofu. (NEC and Hitachi backed out  of the $1.2bn Project Keisoku in May 2009, saying they would lose money on the deal if they proceeded.)
Fujitsu started building what is now called the K supercomputer back in September . The prototype of the K super is not powerful enough to make the Top 500 list, but using 2 GHz Sparc64-VIIIfx processors the machine delivers 828.7 megaflops per watt and that is better than a slew of supers based on IBM's PowerXCell 8i processors, which have dominated the Green500 list for several years.
Three Cell-based clusters running at the Forschungszentrum Juelich lab in Germany and at the Universities of Regensburg and Wuppertal (also in Germany) deliver 773.4 megaflops per watt and are tied for fifth place on the Green500 list. But these machines are a dead end, because IBM is no longer making Cell processors for use in its BladeCenter blade servers.
Three ceepie-geepie machine round out the top 10 of the Green500 list, with a cluster made by Super Micro for the University of Frankfurt using quad-core Opterons and ATI Radeon graphics cards as GPU co-processors coming in at 740.8 meaflops per watt with the eight rank. Number nine on the Green500 list is a cluster at the Georgia Institute of Technology which pairs two-socket cookie-sheet servers from Hewlett-Packard (using six-core Xeon 5600 processors) with Nvidia GPUs; it is rated at 63.9 teraflops (117th on the Top 500 list) delivers 677.1 megaflops per watt. Number 10 on the Green500 list is a Xeon-C2050 cluster at the National Institute for Environmental Studies in Japan, which ranked 102 on the Top 500 list with 74.8 teraflops of sustained Linpack performance and which delivered 636.4 megaflops per watt.
Energy efficiency continues to improve because of the pressure on supercomputer makers to not only deliver more flops, but to take a lot less energy to do it. In the inaugural Green500 list back in November 2007, all of the top 10 machines on the list were BlueGene/L or BlueGene/P supers with the BlueGene/L machines coming in at 210.6 megaflops per watt and the BlueGene/P machines ranging from 310.9 through 357.2 megaflops per watt. In June 2008, Cell-based machines and the Roadrunner Opteron-Cell hybrid at the Los Alamos National Laboratory took over the top three spots on the Green500 list, jumping ahead of the BlueGene/P boxes and ranging from 437.4 to 488.1 megaflops per watt.
More Cell-based supers were added to the top of the Green500 list, pushing above 536.2 megaflops per watt in the November 2008 list, and in June 2009, nothing much else changed at the top except the Grape-DR FPGA hybrid at the national Astronomical Observatory in Japan came in with 428.9 megaflops per watt.
The prototype Tiahne-1 Xeon-Radeon cluster was the first ceepie-geepie to break the top ten on the Green500 a year ago, with a rating of 379.2 megaflops per watt. The Nebulae ceepie-geepie cluster at the National Supercomputing Center in Shenzhen, China, was number four on the Green500 list from June 2010, at 492.6 megaflops per watt. Interestingly, on the June 2010 list, there were two Xeon-Nvidia GPU hybrids in the top 10 and six machines lashing together x64 processors and Cell co-processors. There were only two plain-vanilla x64-based clusters on the list on the current November Green500 list: IBM's BlueGene/Q prototype and Fujitsu's K prototype. All of the others are based on Cell or x64-GPU hybrids. ®