[Beowulf] recommendations for cluster upgrades

Greg Keller Greg at keller.net
Thu May 14 12:33:05 PDT 2009


Product quality can vary from run to run.... especially on new  
hardware with lot's of new features.

an example of the press on issues is....

I think there was a run or 2 after the initial test chips that had  
early adopters scratching their heads, but it was cleaned up quickly.   
Barcelona was the first AMD 4 Core processor package with integrated  
caching across all 4 cores and the first with 4 Flops per core AMD  
proc so they were making some giant leaps while trying to maintain  
backward compatibility with the socket.  Very ambitious, risky, and a  
lot of "cool" factor features... The fact that it worked at all is  
impressive in a certain way!  It didn't come out in a timely fashion  
and the "cool" features didn't really benefit my customer's codes....  
but it certainly created opportunity for coders to revisit what they  
would do with tightly coupled caches between cores.

I'm not a programmer, so I don't really know the short term value of  
some of those features but goodness can only come from more options  
when it comes to the challenge of parallelizing tasks.


On May 13, 2009, at 6:22 PM, Rahul Nabar wrote:

> On Tue, May 12, 2009 at 7:05 PM, Greg Keller <Greg at keller.net> wrote:
>> AMD Barcelona was the first 4 flops per cycle processor from AMD,  
>> and it hit
>> the street with some problems right when the list was coming out in  
>> end of
>> 2007.
> That's interesting. What kind of "problems"? Do CPU designers mess up
> and leave bugs on too?  I heard of an old Intel floating point error
> but nothing else. Do later versions of CPUs get these bugfixes? Or are
> CPU designs pretty static until they come out with a new proc number /
> proc line?
> It might change my perspective on the risks of going for a "brand  
> new" CPU.
> -- 
> Rahul

More information about the Beowulf mailing list