Fact: The annual ACM Gordon Bell Prize is about to be awarded at “SuperComputing 08” (or SC08) which takes place November 15-21 at the Austin Convention Center in Texas. The convention is “the” international conference for high-performance computing, networking, storage, and analysis. The Association for Computing Machinery makes the Bell award “to recognize outstanding achievement in high-performance computing,” in honor of Microsoft’s legendary Gordon Bell, one of the pioneers of supercomputing.
Analysis: Microsoft will be well represented at SC08 (for example, a half-day session on “Windows HPC Server 2008, a Programmers Perspective”) and two MSFT guys I know (Dan Reed and Burton Smith) are on the conference Steering Committee… and one of the main speakers is David Patterson, former ACM president and former chair of Berkeley’s Computer Science department, now director of both the Reliable Adaptive Distributed Systems Lab and the Parallel Computing Lab at UC Berkeley, which is co-funded by Intel and Microsoft.
However I’m just a tiny bit disappointed that no Microsoft Research projects are in line for the Bell Prize this year. Shut out again, same as the last fifteen years! This is hard-core stuff actually, as you can see from the finalists for this year’s ACMGB Prize:
- “Multi-Teraflops Simulations of Disorder Effects on the Transition Temperature of the High Tc Superconducting Cuprates” by Gonzalo Alvarez and a team from Oak Ridge National Laboratory and Cray Inc.
- “0.374 Pflop/s Trillion-particle Particle-in-cell Modeling of Laser Plasma Interactions on Roadrunner” by Kevin J. Bowers and a team from Los Alamos National Laboratory.
- “Scalable adaptive Mantle Convection Simulation on Petascale Supercomputers” by Carsten Burstedde and a team from University of Texas at Austin, California Institute of Technology, and University of Colorado
- “High-frequency Simulations of Global Seismic Wave Propagation using SPECFEM3D_GLOBE on 62K Processors” by Laura Carrington and a team from University of California, San Diego and University of Pau
- “350-450 Tflops Molecular Dynamics Simulations on the Roadrunner General-purpose Heterogeneous Supercomputer” by Sriram Swaminarayan and a team from Los Alamos National Laboratory
- “Linear Scaling Divide-and-conquer Electronic Structure Calculations for Thousand Atom Nanostructures” by Lin-Wang Wang and a team from Lawrence Berkeley National Laboratory
Speaking of teraflops, I spent last week at the GEOINT conference in Nashville, and will be publishing a more substantive post on some of that conference’s content (speakers and panels). But I also wandered down to the exhibit floor a few times, of course, and actually made my way (as is my wont) past every single booth, at a minimum glancing at the products and in many cases getting the full pitch.
True to my oddball form I was most taken not with the geospatial-intelligence software and solutions on display but with a piece of hardware: the Cray CX1 used in the Microsoft booth as a platform to show its potential for high-end geospatial uses running Windows High-Performance Computing Server 2008. I snapped the accompanying photos with my cellphone, so they’re not the best, but the Cray website page about the machine has better photos if that’s what you’re looking for.
What you should be looking for is performance, and they bill this box as “for individuals and departmental workgroups who want to harness HPC without the complexity of traditional clusters.” For the GEOINT crowd, that translated (as the Windows HPC Server sign says in the photo) to computational support for Modeling and SImulation, Signal Processing, Chemical/Biological Defense, and “Electronic Battlefield Environments” (think realtime Future Combat Systems for example).
The specs support the HPC label, even though the box is not much bigger than the case of the IBM PC-XT clone which I bought from a Berkeley sweatshop back in 1985. Here are some impressive highlights:
- (up to) 16 Intel Xeon quad-core processors
- (up to) eight blades per chassis (deskside or rackmount)
- (up to) 32 gigs of memory per blade
- (up to) four high performance NVIDIA Tesla GPU computing processors
- runs on standard office power!
Interestingly, in this time of enthrallment over Cloud Computing and using someone else’s megadatacenter, Cray makes an interesting middle-ground pitch for the CX1: “If your datacenter is currently tapped out in either available power or cooling (as many of them are), a CX1 in an office will help to distribute the power and cooling load away from the datacenter. The CX1 is designed to be utilized in an office environment and can be easily connected to your network for easy access by users.”
And one Microsoft partner was showing at GEOINT their specialized CX1 version for deployable environments, Criterion’s “Cube C15” which has been benchmarked to750 Gigaflops of computing power but needs no extra cooling power – it runs at room temperature.
So, forget the cloud, and put cumulo-nimbus power in your office? For some scenarios, might be worth a shot.
Oh, by the way, in the photo, the little pile of stickers on the table beside the box read, “My other computer is a Cray.” I have one on my laptop now….
Filed under: Government, innovation, Intelligence, Microsoft, R&D, Technology | Tagged: ACM, Austin, Berkeley, biological defense, chembio, chemical defense, CIT, cloud, cloud computing, compsci, computer, computer science, computing, Cray, Criterion, geoint, geospatial, Gordon Bell, GPU, HPC, HPC Server, intel, Intelligence, IT, LANL, laser, Los Alamos, Microsoft, Microsoft Research, modeling, Nashville, NVIDIA, Oak Ridge, ORNL, plasma, quadcore, SC08, server, signal processing, simulation, supercomputer, supercomputing, tech, Technology, Tesla, UCB, UCSD, Windows, Xeon |