Get to know me!
|
by the SCD consultants
Q: What did the SCD consultant say when they unplugged shavano?
A: Hip, hip, ouray!
-- NCAR user Dr. Robert Schlesinger
Introducing ouray
W
hen shavano, the Y-MP8/864, was decommissioned in
June 1997 due to an aging architecture, disk problems, and
maintenance costs, SCD consultants invited users to move their
codes to ouray.
Ouray is a CRAY J90se computer with 24 CPUs and 1 billion words of central memory. It has 16 times as much memory as shavano, and although its processor speed is about one-half the speed of a Y-MP processor, it sustains up to twice the gigaflops as shavano due to its increased number of processors and vastly increased memory. In fact, with 44 million words of memory per CPU, ouray is ideally suited to run the jobs of NCAR community users, which are often single tasked.
Ouray passed acceptance testing and became openly available to the general user community on 12 May 1997. It had been initially ordered with 20 CPUs, but SGI/Cray offered NCAR an additional 4 CPUs at no cost in compensation for the long and laborious checkout of the system. The additional CPUs were installed in the system on 27 May 1997.
J90 vs J90se
The
J90se is a new model of the J90-series Parallel Vector Processor (PVP) product from SGI/Cray. The original J90 systems (such as aztec and paiute) are now referred to as "J90 classics." The J90 classic systems clock their scalar and vector processors at 100 MHz and the theoretical peak performance is 200 megaflops per CPU. The 90se system clocks its vector processor at 100 MHz and its scalar processor at 200 MHz.
Though the theoretical peak performance of a J90se processor remains at 200 MHz, because of its double-speed scalar processor, the delivered performance on typical "real world" codes is improved over the J90 classic systems. The faster scalar processing of ouray, as with its large memory-to-CPU ratio, makes it well suited to run the (often scalar) job mix of NCAR community users.
To date, SCD is quite pleased with ouray's overall performance. Though the per-processor performance of a systme is highly dependent upon the applications being run, we presently feel that the J90se is demonstrating its viability as a real workhorse in the NCAR computational environment.
Comparing ouray and paiute
T here
are several advantages to using ouray over paiute:
- Ouray has four times as much memory as paiute.
- Ouray has 24 CPUs, while paiute has 16.
- Ouray is somewhat faster than paiute at scalar arithmetic (the "se" in J90se
means "scalar enhanced"). Ouray's scalar processor runs at 200 MHz, while
paiute's scalar processors run at 100 MHz. Cray Research claims about a 20% improvement for normal, mixed scalar/vector codes.
Comparing ouray and shavano
Users
will note two major differences from shavano:
- Ouray (like paiute) has no SSD.
- As with all new Cray computers (manufactured since September 1996), the CF77/CFT77 compilers are no longer available, so Fortran users will need to switch to Fortran 90 if they haven't yet done so.
For more information on converting to F90, see:
If they haven't yet done so, users will also need to set up their environment using "modules"; for instructions, see:
Down home with ouray
For more information on the J90se, see the ouray home page:
www.scd.ucar.edu/computers/ouray/
|
