🚀 go-pugleaf

In article <BxKI38.DM7.2@cs.cmu.edu>, lindsay+@cs.cmu.edu (Donald Lindsay) writes:
>>
>>System                100x100 dp      1000x1000 dp
>>
>Compaq Deskpro 486/33	      1.4
>DEC 5000/200 (25 MHz R3000)   3.7
>Cray-1			     11		       31
>>DEC 3000/400S AXP           26.4              70.8
>>DEC 10000/610 AXP           42.5             111.6
>Convex C3810 (1 proc)	     44		      113
>Cray C90     (1 proc)	    387		      874

      (NOTE: The performance above which is attributed to a CRAY-1 is
      actually that of a CRAY S-MP, NOT a CRAY-1.  The CRAY S-MP is
      SPARC based system with scalar and with vector processors.  There
      was also a product announcement last week from Cray Research
      Superservers, the former FPS, regarding a new vector CPU for the
      CRAY S-MP.)

We must be careful when comparing new architectures with old
architectures because of the effect of software.  The CRAY-1 has two
entries in the LINPACK benchmark report:

System                 100x100 dp    1000x1000 dp
CRAY-1S  cf77 2.1        27 Mflop/s     110 Mflop/s
CRAY-1S  CFT 1.12        12             110

That CFT 1.12 results was generated in the early 80s (1983?) with cft
1.12.  The cf77 2.1 result was generated in 1989.  With current
compilers from Cray Research, the 100x100 performance for the CRAY-1S
would be approximately 35 or 40 Mflop/s.  On this particular benchmark,
we would see a 3x speed improvement attributable to software.  (The
1000x1000 performance should not change with time because it usually
generated with "hand-optimized" libraries.)

The improvement of microprocessor speed of the past decade has been
tremendous.  How much of this improvement is attributable to software?

Charles Grassl
Cray Research, Inc.
Eagan, Minnesota  USA