® ®
1. Accelerating Applications
- Industry-leading MPI results - At 1microsecond, SGI's NUMAlink™ interconnect technology offers
industry-leadership in MPI latency and bandwidth (6.4GB/s bi-directional per link), delivering great performance on MPI
codes. A good example is SGI Altix leading performance on the most commonly used mesoscale weather forecasting codes in
the world: WRF and MM5. (Reference: http://www.mmm.ucar.edu/mm5/mpp/helpdesk/20040304a.html)
- Scaling to 512P and beyond at NASA - NASA has achieved extraordinary scaling of commonly used applications with
minimal effort due in large part to what they describe as the 'trivial' nature of load-balancing on their 512 processor global
shared memory Altix systems. (Reference: http://www.nas.nasa.gov/About/Projects/projects.html). This model has reduced programming time for scaling and optimization of very large, complex applications from many
man-years down to a few weeks. The 512P results have rivaled and even exceeded those of the $500M Earth Simulator.
- Bioinformatics - A UK-based pharmaceutical research institute has
developed a genomics search and matching algorithm that produces results up to
1000x faster than the commonly used BLAST application. This algorithm relies
on 192GB of memory available on each of two SGI Altix servers with 4 and 16
processors.
2. Scaling Problem Size
- NWChem - The popular code NWChem is routinely run at one SGI customer site on only 4 processors of an SGI Altix but using 80% of the 2 Tbytes of globally shared memory configured in that system. While execution may not scale efficiently across more than a modest handful of processors, big gains can be had simply by exploiting additional memory capacity.
- Nastran - A similar example is Nastran, which normally does not scale beyond a few processors. This code sees considerable speedup when the job has access to larger shared memory.
3. Real-time Interaction with Massive Datasets
- Large-scale seismic dataset interaction - Marathon Oil has met the challenge to
deliver larger and larger seismic data to the interpreting geoscientist. For the first
time, geoscientists can visualize and interact with more than 400GB of seismic data in real
time. This is more than 4 times the previous record. (Reference: http://www.lgc.com/)
4. Unprecedented Computational Options
- Massive structural analysis - Recently, ANSYS announced that it became the first
engineering simulation company to solve a structural analysis model with more than 100
million degrees of freedom (DOF), making it possible for ANSYS customers to solve full
resolution models of aircraft engines, automobiles, construction equipment and other
complete systems. In a joint effort with Silicon Graphics, Inc. (SGI), the 111 million DOF
structural analysis problem was completed in only a few hours using only a handful of
processors on a large memory SGI Altix computer. As a result of this work, ANSYS entered
into a three-year partnership with SGI to advance the capabilities of ANSYS in parallel
processing and large memory solutions.
- Rapid response data analysis - A US Government Agency is using 32 processor SGI Altix
systems equipped with 4 TB of memory as the only solution allowing them to handle the
massive data mining and search algorithms essential to their work.
- Memory resident databases - Xcelerix IMDB and database accelerator has demonstrated orders of magnitude performance gains over traditional, disk resident databases on SGI Altix systems for databases containing up to 500 Million records (120 GB).
