[geeks] New Itanium machines from SGI

[Dave McGuire]

On Wednesday, January 8, 2003, at 01:07 PM, Mike Meredith wrote:
>>> I wouldn't be surprised if it is true ... MIPS is hardly a speed
>>> demon anymore.
>>
>>    Umm, compared to what, exactly?  These are the numbers that I see.
>>
>> proc			clock	SPECint	/cyc	SPECfp	/cyc
>> -----------------------------------------------------
>> R12000			 400	353		0.882	407		1.017
>> R14000			 500	427		0.854	463		0.926
>> PA-8700			 750	604		0.805	576		0.768
>
> Don't those figures show that MIPS whilst the most 'efficient', isn't
> the fastest ? Or am I missing something ?

   You're missing something. ;)  This is an old SPECmark table, and MIPS 
chips have upped their clock rates since then.  So have other 
processors, but the gap is narrowing.

   So, let's say you blow a metric buttload of money on a computer room 
full of <processor xyz> and invest a bunch of money (perhaps an 
imperial buttload) in software development for those processors.  When 
process technology hits the laws-of-physics-imposed barrier of ~4GHz or 
so (by the last estimates I read) who's gonna be faster?

   Also, if performance is the name of the game, then SMP is king.  
Which ones scale better in SMP configurations?

   When speaking of cheap uniprocessor machines, sure, MIPS-based boxes 
aren't the fastest machines that you can pick up the phone and order 
today.  That probably won't change by the end of this month, and maybe 
not by the end of the year.  But I wouldn't bet on it for any 
significantly longer period of time.  When you consider the fact that 
the "buy a new computer every six months" days have been over for a 
couple of years now, this becomes significant.

> I certainly wasn't aiming to say that MIPS are dog slow, but that they
> aren't the fastest possible CPU you can buy.

   Understood.

         -Dave

--
Dave McGuire           "She's a cheek pincher.  I have scars."
St. Petersburg, FL                          -Gary Nichols