kalra@cit-vax.UUCP (06/14/87)
Could someone please point to a reference to various benchmarks, specifically the Dhrystones. Also what does the Norton Index mean. Thanks
paul@cgh.UUCP (06/15/87)
In article <3019@cit-vax.Caltech.Edu> kalra@cit-vax.UUCP (Devendra Kalra) writes: > >... Also what does the Norton Index mean. One of the programs distributed with the Norton Utilities is SI, or Speed Index. When run, it produces a number which is supposed to be an Index relative to the original IBM PC. That is, a Norton SI of 2.0 is supposed to mean that the current machine is 2 times faster than a standard PC. However, beware: the test uses a small subset of machine instructions and the values it gives can vary wildly when used on different families of processors. It gives entirely misleading results when comparing say, an 8088 and an 80286, or an 8086 and a V20. To summarize: the Norton Index means nothing. You would be well advised to ignore all claims based on it unless you are comparing like processors, in which case it can show you the relative clock speeds.
NPASMITH@TUCCVM.BITNET (Philip A. Smith) (11/13/87)
Machine | CPU |N=30 |N=30 |N=50 |N=50 | N=70 || | | | |FORT | C |FORT | C | FORT ||FORT | C | IBM XT | 8088/7 |-----|60.57|-----|-----|------||-----|62.93| Zenith Z158 | V20/8087 |50.55|39.31|-----|-----|------||-----|42.96| Everex 1800 | 80286 |-----|299.0|-----|-----|------||-----|17.18| IBM AT 6MHz | 80286/7 |76.89|60.26|-----|-----|------||-----|23.05| IBM AT 16MHz| 80386/7 |15.73|13.60|-----|61.60|------||-----| 7.48| Zenith Z386 | 80386 |-----|132.2|-----|-----|------||-----| 6.85| | Sun 3/60 | 6.43|-----|35.95|-----|100.21||-----| 9.76| * | Microvax |12.37|-----|42.05|-----|108.35||15.35|-----| * | Vax11/785| 6.66|-----|20.13|-----| 48.88||11.18|-----| * | IBM 4341 | 4.16| 7.26|19.03|33.54| 52.09|| 8.45| 6.65| * | IBM 4381 | 1.65| 2.90| 7.10|13.34| 19.55|| 4.23| 3.73| * | IBM 3081 | 0.56| 1.01| 2.53| 4.71| 6.93|| 1.06| 1.28| * | IBM 3090 | 0.18| 0.39| 0.80| 1.80| 2.19|| 0.51| 0.69| +----------+-----+-----+-----+-----+------++-----+-----+ * CPU time reported by operating system, all others elapsed time. Interesting Observations: The Everex 1800 and Zenith Z386 did not have math coprocessors (I would expect them to do very well if they did) ! The "16MHz IBM AT" was a 6MHz AT with an Intel Inboard/386 card and 80387 math coprocessor. Look at the integer time for the Z386! This was a base model without the 64k memory cache, so this time could be improved. Something weird is happening with the Sun 3/60 -- for longer runs it is "asymptotically" approaching Microvax performance. The Vaxes integer performance was disappointing. This ought to make owners of XT/AT/386's feel better. I was going to have a "cost" column for each machine but I don't have the slightest idea what a 3090 , etc. costs (10**7 $ ?), plus university pricing is vastly different from retail. If someone wants to enlighten us, please do. When you consider that many of the big machines run with an "overload factor" (elapsed time/CPU time) of 5-20, you might get your work done faster on a micro. If you've used TSO on an overloaded big IBM, you won't get too enthused about "multiuser" PCs. ************************************ Floating-point benchmark - triple integration by the mid-point rule. Note that the exp(x/y) can be pulled out of the inner loop by a smart optimizing compiler -- if a machine has such a compiler then it deserves the better benchmark time. (Not efficient but easy to type into machines that aren't networked to anything.) *** bench.c *** #include <stdio.h> #include <math.h> main() { double d,start,x,y,z,v=0; int i,j,k,n; puts("enter n:"); scanf("%d",&n); d=1.0/n; start=1.0+0.5*d; for( x=start,i=0; i<n; x+=d, ++i) for( y=start,j=0; j<n; y+=d, ++j) for( z=start,k=0; k<n; z+=d, ++k) v += sin(x*y*z)*exp(x/y)*log(y/z); v *= d*d*d; printf("\nn=%d, v=%25.16le",n,v); } *** bench.for *** IMPLICIT REAL*8(A-H,O-Z) WRITE(*,*) 'ENTER N:' READ(*,*) N D=1.0/N START = 1.0D0 - 0.5*D X=START DO 10 I=1,N X=X+D Y=START DO 10 J=1,N Y=Y+D Z=START DO 10 K=1,N Z=Z+D 10 V=V+ DSIN(X*Y*Z)*DEXP(X/Y)*DLOG(Y/Z) V=V*(D**3) WRITE(*,*) 'N=',N,' V=',V STOP END ************************************ Integer benchmark -- Calculate 1899 prime numbers 100 times. (One should check that an optimizing compiler does not move the body of the program outside the "100 loop" [change the 100 to 50 and see if execution time is halved.]) (Adaptations of sieve of Eratosthenes benchmark (Byte magazine orig.)). *** sieve.c *** #define S 8190 char f[S+1]; main() { register int i,p,k,c,n; for (n = 1; n <= 100; n++) { c = 0; for (i = 0; i <= S; i++) f[i] = 1; for (i = 0; i <= S; i++) { if (f[i]) { p = i + i + 3; k = i + p; while (k <= S) { f[k] = 0; k += p; } c++; } } } printf("\n%d primes.\n", c); } *** sieve.for *** INTEGER I,P,K,C,N,F(0:8190) DO 10 N=1,100 C = 0 DO 20 I=0,8190 20 F(I) = 1 DO 30 I=0,8190 IF (F(I).NE.0) THEN P = I + I + 3 K = I + P 50 IF (K.GT.8190) GOTO 40 F(K) = 0 K = K + P GOTO 50 40 CONTINUE C = C + 1 END IF 30 CONTINUE 10 CONTINUE WRITE(*,*) C, 'PRIMES' STOP END ***** End of file ******
promac@iisat.UUCP (Promac Systems) (05/30/89)
After scanning through all the latest "CLONE" mags looking at all of the benchmarks(I know that they are not the end of all comparsions), I thought that I would like to see for myself how the Motorola based systems standup. Does anyone have the C source code for the LinPack, matrix, and other assorted benchmarks that they could E-mail to me. later Barry Comer promac@iisat.UUCP
conan@vax1.acs.udel.EDU (Robert B Carroll) (05/31/89)
In article <14@iisat.UUCP> promac@iisat.UUCP (Promac Systems) writes: >Does anyone have the C source code for the LinPack, matrix, and >other assorted benchmarks that they could E-mail to me. >Barry Comer >promac@iisat.UUCP look in uunet. i remember getting some dhrystone and whetstone stuff. -- conan@vax1.acs.udel.edu OR conan@192.5.57.1 CONAN THE BARBARIAN of Cimmeria
ecf_udwo@jhunix.HCF.JHU.EDU (Douglas W O'Neal) (05/31/89)
>Does anyone have the C source code for the LinPack, matrix, and >other assorted benchmarks that they could E-mail to me. >Barry Comer >promac@iisat.UUCP You can get these from from the netlib. Send a mail message consisting of the line send index to netlib@anl-mcs.arpa to get the full list of software available. Doug O'Neal