[KLUG Hardware] AMD releases preliminary x86-64 Opteron SPEC numbers ...

[Bryan J. Smith]

According to the article below, AMD has released prelimintary SPECint and SPECfp
numbers that only test the computational power of the processor.  For those of
us that have been following AMD's HyperTransport exploits, everyone knows the
Opteron's "bonus" is in its on-die "northbridge" with an integrated memory
controller and HyperTransport interconnect upto 8-way.  So these numbers don't
fully reflect the excellent system design, just the chip.

Although the numbers are "unofficial" since the hardware hasn't been released.

SPECint Quote:  
   "An Opteron running at 2GHz with PC2700 DDR SDRAM achieved
    an unofficial score of 1,202 on SPECint, well above the 
    957 posted for Intel's 2.8GHz Xeon and 810 reported for
    its 1GHz Itanium 2. Both tests were run on 32-bit
    applications, since 64-bit applications for Opteron
    are not available."

That is just with 32-bit numbers!  It whips the P4 Xeon silly, which has
_better_ SPECint numbers than the Itanium2.  Considering the P4 beats the Athlon
heavily at SPECint, thanx to Intel's optimizing compilers (but not so with the
GNU ones), but the Athlon is pretty damn competitive with the P4, this is
excellent news.

SPECfp Quote:
   "AMD's unofficial score for Opteron on SPECfp was also strong,
    achieving a result of 1,170, compared with 887 for the 2.8GHz
    Xeon, but well short of the 1,370 achieved by the 1GHz
    Itanium 2.  However, both Opteron scores were higher than
    those posted by other major 64-bit processors, including
    IBM's Power4 and Sun Microsystems Inc.'s UltraSparc III."

Likewise, those 32-bit SPECfp numbers look promising, and could be better when
64-bit code is run (because SPECfp is 64-bit double precision).  RISC has always
been the domain of high SPECfp numbers, and the most anal of RISC designs, the
Digital (now Compaq/HP) Alpha 264 has numbers almost as high as the Itanium2
from 4 years ago at only 667MHz!

eWeek:  "AMD Touts Opteron Performance"  
  http://www.eweek.com/article2/0,3959,636049,00.asp  

-- 
Bryan J. Smith, E.I.            Contact Info:  http://thebs.org
A+/i-Net+/Linux+/Network+/Server+ CCNA CIWA CNA SCSA/SCWSE/SCNA
---------------------------------------------------------------
           limit      guilt   =     { psychopath,
         remorse->0                    innocent }