CRAY computers

In the mid-1990s, virtually all scientific computation will be able to access an international network including advanced multi-processor supercomputers having two orders of magnitude more available computer power than today s largest Cray computers. The PC workstation itself of that time will have practical capabilities equivalent to a Cray 1 (extrapolate forward two years from Table 10.2 ). 

The author acknowledges fruitful collaborations with several of his coworkers, particularly. Dr. K.K. Datta and Dr. S.K. Bhattacharya. This work has been supported in part by the United States Department of Energy (Division of Chemical Sciences) and by the Alfred P. Sloan Foundation. The author Is grateful to the United Telecom Computing Group (Kansas City) for generous support of the CRAY computer time. 

The most time-consuming step in the Knowles and Handy determinantal FCI scheme, as noted above, is a series of matrix multiplications, so the code performs very efficiently on Cray computers. Calculations as large as... 

S. Hunsicker, for discussions and collaboration on work described here. The calculations were performed on Cray computers of the Forschungszentrum Julich and the German Supercomputer Centre (HLRZ). 

R. Emenwein, M. Benard, and I. Shavitt, Vectorizing a Sequence of Conditional Branches the Calculation of the Class Index of Two-Electron Repulsion Integrals on Cray Computers, Comput. Phys. Com-mun. 48, 175-180 (1988). 

This research was supported by a grant from the US Department of Energy (Grant No. FG05-86ER45234). Calculations were performed on CRAY computers at the Supercomputer Center at Livermore, California, and the Georgia Institute of Technology Center for Computational Materials Science. 

The Cray computer series continued in development, and the successors of the Cray-1 design were for more than 25 years the fastest supercomputers in the world. 

Table 1 E2LP board peripherals that are memory mapped on the Cray computer...

This completes the outline of FAMUSAMM. The algorithm has been implemented in the MD simulation program EGO VIII [48] in a sequential and a parallelized version the latter has been implemented and tested on a number of distributed memory parallel computers, e.g., IBM SP2, Cray T3E, Parsytec CC and ethernet-linked workstation clusters running PVM or MPI. 

The Fourier sum, involving the three dimensional FFT, does not currently run efficiently on more than perhaps eight processors in a network-of-workstations environment. On a more tightly coupled machine such as the Cray T3D/T3E, we obtain reasonable efficiency on 16 processors, as shown in Fig. 5. Our initial production implementation was targeted for a small workstation cluster, so we only parallelized the real-space part, relegating the Fourier component to serial evaluation on the master processor. By Amdahl s principle, the 16% of the work attributable to the serially computed Fourier sum limits our potential speedup on 8 processors to 6.25, a number we are able to approach quite closely. 

D. Fincham and B. J. Ralston, Molecular dynamics simulation using the Cray-1 vector processing computer , Comp. Phys. Comm., Vol 23, no 2, 127-34, 1981. 

Computer platforms where 1 represents IBM PCs and compatibles, 386 or better 2, Unix-based work stations (or equivalent) including HP/ApoUo, SUN, and IBM RS6000 3, DEC VMS 4, IBM mainframe, MVS or VM, or compatible and 5, others including Cray, Data Gen., etc. 

Supercomputers from vendors such as Cray, NEC, and Eujitsu typically consist of between one and eight processors in a shared memory architecture. Peak vector speeds of over 1 GELOP (1000 MELOPS) per processor are now available. Main memories of 1 gigabyte (1000 megabytes) and more are also available. If multiple processors can be tied together to simultaneously work on one problem, substantially greater peak speeds are available. This situation will be further examined in the section on parallel computers. 

The most commercially successful of these systems has been the Convex series of computers. Ironically, these are traditional vector machines, with one to four processors and shared memory. Their Craylike characteristics were always a strong selling point. Interestingly, SCS, which marketed a minisupercomputer that was fully binary compatible with Cray, went out of business. Marketing appears to have played as much a role here as the inherent merits of the underlying architecture. 

Logic Analysts version of SETS operate on Cray X-MP, Y-MP, CDC 930, IBM 3090, all 80386 and higher personal computers, IBM RS-6000, Sun SPARC 1, DEC VAX, and HP-700 computers. WAMBAM uses the original GO algorithm of table hookup and does not provide ihe intermediate step of finding the minimal cutset (current versions of GO provide cutsets ). 

Path integral Monte Carlo simulations were performed [175] for the system with Hamiltonian (Eq. (25)) for uj = ujq/J = A (where / = 1) with N = 256 particles and a Trotter dimension P = 64 chosen to achieve good computer performance. It turned out that only data with noise of less than 0.1% led to statistically reliable results, which were only possible to obtain with about 10 MC steps. The whole study took approximately 5000 CPU hours on a CRAY YMP. 

O.P. acknowledges support of the Ministry of Science and Technology (CONACyT) of Mexico (Grant No. 25301-E) and the National University of Mexico (project INI 11597). O.P. is also grateful to Silicon Graphics Inc.-Cray Research of Mexico for partial financial support and generous allocation of computer time. 

The performance data for C5H12 is more representative of real-world problems (although this system is still on the small side). These jobs were all run on a Cray T-94 computer system ... 

The authors gratefully acknowledge Cray Research for providing the computer time to complete this study and thank John Carpenter and Carlos Sosa for their assistance. 

John Carpenter and Carlos Sosa of Cray Research generously provided the computer time and technical assistance for the resource use study in Chapter 6. 

---