Hyperthreading in P4

Rocky McGaugh rocky at atipa.com
Mon Apr 29 16:09:08 PDT 2002


On Mon, 29 Apr 2002, Steve Gaudet wrote:

> 
> Key issues are:
> 1.  Code must be threaded.
> 2.  BIOS and O.S. must be enabled.
> 	- RH has a patch available on their site.
> 
> As for the performance vs PIII, I strongly recommend that application
> developers use our C and Fortran compilers for Linux.  GCC is not
> well-optimized for Netburst architecture, PGI is o.k., and our compilers
> really fly!  In addition, our compilers generate the best code for PIII,
> AMD, and P4P/Xeon based systems so you really can't lose.
> 
> Pls see the following Hyper-Threading whitepaper (preliminary) for all the
> gory details about req'ts.> 
> 
> 

Fine and dandy. The problem is with no way to bind processes to 
processors, it's quite easy for 2 heavy processes (or threads) to migrate 
to the same physical CPU, leaving 2 smaller threads (or processes) on the 
other physical CPU.

For CPU bound apps, its too unpredictable without the processor affinity 
stuff.


-- 
Rocky McGaugh
Atipa Technologies
rocky at atipatechnologies.com
rmcgaugh at atipa.com
1-785-841-9513 x3110
http://1087800222/
perl -e 'print unpack(u, ".=W=W+F%T:7\!A+F-O;0H`");'




More information about the Beowulf mailing list