Saavedra-Barrera, R. H. (1990). "The SPEC and perfect club benchmarks: promises and limitations." Hot Chips Symposium 2. Santa Clara, California.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....stream) and a separate execution time for each is obtained. An average of these times may then serve as the performance rating for the system (other scores are possible) Several currently popular benchmarks use this methodology, including the PERFECT (Cybenko 1990,Grassl and Schwarzmeier 1990, Saavedra Barrera 1990) and SLALOM (Gustafson et al. 1990) benchmarks. A third example is the SPEC benchmark suite (Saavedra Barrera 1990) The Standard Performance Evaluation Corporation (SPEC) is a nonprofit corporation formed to establish, maintain and endorse a standardized set of relevant benchmarks that can be ....

....rating for the system (other scores are possible) Several currently popular benchmarks use this methodology, including the PERFECT (Cybenko 1990,Grassl and Schwarzmeier 1990, Saavedra Barrera 1990) and SLALOM (Gustafson et al. 1990) benchmarks. A third example is the SPEC benchmark suite (Saavedra Barrera 1990). The Standard Performance Evaluation Corporation (SPEC) is a nonprofit corporation formed to establish, maintain and endorse a standardized set of relevant benchmarks that can be applied to the newest generation of highperformance computers (quoted from SPEC s bylaws) SPEC 1997) Although SPEC ....

Saavedra-Barrera, R. H. (1990). "The SPEC and perfect club benchmarks: promises and limitations." Hot Chips Symposium 2. Santa Clara, California.

....in the machine. The IBM XLF compiler is capable of eliminating most of the loads by reusing registers and applying predictive commoning. 5.1.4. MATRIX300 It has been documented that this benchmark is completely dominated by a single basic block, which accounts for 99 of the execution time [Saav90a, 90b, 92a]. This basic block implements the SAXPY vector to vector operation Y[1,1:N] Y[1,1:N] A X[1,1:N] This subroutine is used in the program to compute eight different variations of matrix multiplication, each representing a particular operation between the three matrices and their transposes. ....

Saavedra-Barrera, R.H. "The SPEC and Perfect Club Benchmarks: Promises and Limitations ", Hot Chips Symposium 2. Santa Clara, CA, August 1990.

....2.37 3.06 13.81 4.51 1.90 nasa7 2.10 3.01 14.95 4.97 2.37 fpppp 3.88 1.45 15.16 10.45 2.70 tomcatv 2.86 2.13 17.45 8.18 2.86 doduc 2.65 1.67 13.16 7.85 2.96 espresso 1.70 1.06 5.40 5.09 2.99 eqntott 1.08 1.25 4.38 3.51 3.25 li 1.62 1.10 6.53 5.97 3.69 geo. mean 2.17 1.71 9.87 5.77 2. 66 benchmark [28]. We were not able to measure gcc on our instrumented VAX 8700, so all of our results are for the nine other benchmarks only. Also, our run of espresso used just one of the four input circuits (bca) We used the most up to date versions of compilers that were available to us in mid 1990 on both ....

....VAX, which are handled by costly microcode traps in the 8700. Spice has the lowest RISC factor, and its own pecularities. Saavedra Barrera has observed that the particular input circuit used in the SPEC version causes spice to spend an unusually large amount of time in one small integer routine [28]. We have already seen that spice has the highest Dstream miss ratios, with the MIPS value being quite high (26.9 percent) It also has the lowest number of loads per instruction on MIPS, so that the high miss ratio hurts less than it otherwise might. 5 Conclusion We now speculate briefly on ....

Saavedra-Barrera, R.H. The SPEC and Perfect Club Benchmarks: Promises and Limitations. Hot Chips Symposium 2. Santa Clara, CA, Aug. 1990.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC