[rescue] Sun Kit Needed for EE Student Here

["Javi Mahai <lefa at ucsc.edu>" at ucsc.edu]

On Tue, 2 May 2006 21:53:16 -0500 (CDT)

> Those are some awfully specific numbers for a very vague 
>operation.  So,
> since that's true, do you imply that -every- Cadence 
>operation takes 10
> minutes on a 2.2GHz Pentium or that only those which 
>take 10 minutes
> take 29 minutes on a V880, regardless of how they may 
>benefit from each
> system's different characteristics?  That, of course, 
>also says nothing
> about the memory load involved, storage I/O, number of 
>iterations
> performed, or any other number of variables.

I do EDA for a living, for the most part, an off the shelf 
opteron/Xeon64 will kick the arse and take the name of an 
USIV in raw processing horsepower. Doing verilog 
simulations and synthesis (Synopsys) on our systems we 
have a speedup factor of over 3 when running on a 2.2Ghz 
opteron (SUN Ultra40) vs a late model USIV machine. The 
only upside to the SPARC machine in our lab is that it 
holds more RAM (16GB) so some synth jobs are faster 
because we go over the 8GB we have in the Opteron 
machines. For every other CAD operation, the Opteron 
machines is significantly faster. The AMD machine is not 
almost twice as fast clockwise, it it also out-of-order. 
It would be interesting to compare an Opteron with an 
actualy CPU contender in the SPARC side, but I do not have 
access to any SPARC64 par. Sparc machines are only of use 
to us for jobs that have more memory footprint than what 
can be fit on the Operon parts that we have, and that is 
the niche SUN plays in when it comes to the EDA market. 
Everyone is moving in droves to X86-64 when it comes to 
desktop EDA.

> So, because a particular PC will run a particular 
>operation in Cadence
> in nearly 1/3 the time that a particular Sun will run a 
>particular
> operation in Cadence (same one?  We don't know!), the 
>whole SPARC
> platform is dog slow?  Nice.  That certainly makes 
>performance
> evaluations a lot easier.  

Not really but for a single thread, for the most part, a 
SPARC from sun is dog slow when compared to late model X86 
offerings. Problem is that most EDA code out there is 
poorly threaded, so even sun is offering X86 offerings 
when it comes to raw horsepower development machines on 
the desktop. Things may change when/if SUN gets Fujitsu's 
out-of-order parts... but as of right now, SPARC from SUN 
is dog slow on a single thread.