Torbjorn Granlund
13 Mar 2003 12:02:09 +0100

Sanjeev verma <> writes:

  I need to enquire about the bench marks of Sun Blade 2000 900 MHz Copper
  CPU. If this machine has been benchmarked please let me know the bench
  marks. As per Sun this machine carries Copper CPU compared to the
  earlier which was based on aluminium. Please advise me on this.
Are you asking if we've run GMPbench on that processor variant,
or if somebody on this list have benchmarked such a processor
with any benchmark?

I ran GMPbench on a "Sun Fire 15000" with 900MHz processors.  I
cannot see anything about it having copper processors in the
`prtconf' output.

Why does this matter?  Surely, Sun didn't change the processor
pipeline or instruction set.  Perhaps copper allows them to reach
somewhat higher frequencies, but a 900MHz Al UltraSPARC should be
indistinguishable from a 900MHz Cu UltraSPARC, unless your
dissect it.

(What are the reasons for UltraSPARC's terrible GMPbench scores?
A combination of inadequate instruction set and poor
implementation of instructions critical to GMP.  Other current
processors have integer multiply support for generating all
product bits.  UltraSPARC just generates the low 64 bits.  Worse,
these instructions are not pipelined.

In GMP, we use floating-point operations as a partial workaround
for these architectural shortcomings.  If there would be interest
enough to sponsor an UltraSPARC optimization effort, this
floating-point code could be further optimized to gain about 2x
improvement for UltraSPARC 3.)


"spam, spam, spam, egg and spam" -- Monty Python