🚀 go-pugleaf

In article <42dgil$1o6@sloth.swcp.com> Harlan W. Stockman,
hwstock@swcp.com writes:
>marlin@bga.com (Marlin Frederick, Jr.) wrote:
>>ejohnso2@swarthmore.edu (Ellen S. Johnson) wrote:
>>>
>>>SPECint92:  approximately 200    SPECint92:  approximately 133
>>>SPECfp92:   approximately 200    SPECfp92:   approximately 105
>>>
>
>>performance.  Intel is famous for its Intel Xpress Box.  This is nothing but
>>a bogus system designed to maximize the Intel performance numbers for Spec
>>marks.  Really there is a third flaw.  Very few developers do as good a job
>>optimizing thier software for the hardware as the people running these Spec
>>test do.
>
>And IBM (Apply by proxy) is famous for quoting calculated SPECint on 
>hypothetical systems with 2 MB L2 cache with no wait states.  
>
>I'd bet that many inexpensive commercial Pentium systems, based on the 
>Triton chip set and pipeline burst cache and EDO RAM, can now best the 
>performance of the Intel Xpress box.
>

Here are the IBM SPEC92 figures

 Model               40P    42W/T   42W/T   43P-100 43P-120  43P-133
   Technology        PowPC    PowPC   PowPC    PowPC   PowPC    PowPC
   Processor           604      604     604     604     604      604
   MHz                 120      120     120     100     120      133
   I-cache              16       16      16      16      16       16
   D-cache              16       16      16      16      16       16
   Ext cache (MB)     0.25        0     0.5    0.25     0.5      0.5

   Performance(1)
   SPECint92         147.3    118.2   150.2   128.1   157.9    176.4
   SPECfp92          132.9    116.5   146.5   120.2   139.2    156.5

   SPECbase_int92    121.1     95.2   121.8   104.3   127.1    142.2
   SPECbase_fp92     120.7    106.4   133.5   104.8   129.0    146.2


Note that not one of these use a 2MB L2 cache.
IBM systems are considered (by those who really care about memory 
bandwidth) to have the bext memory bandwidth available for workstations 
making them the machines of choice when working on very large arrays or 
on code with very poor data locality.

Maynard Handley

.
.
.
.
.