[Beowulf] dual core intel/amd
diep at xs4all.nl
Thu Apr 28 06:16:10 PDT 2005
Several programs tested under which Diep at sudhian.com
For diep : http://www.sudhian.com/showdocs.cfm?aid=667&pid=2543
If you ask me intel is in serious troubles with respect to beowulfs.
The dual core opteron is just outperforming the dual core P4 so much.
Partly because intel has weakened the weak L1 and L2 caches even more.
Compare the new dual core p4 has a 4 cycle L1 cache which can do 1 read, at
a real tiny cache.
the dual core opteron has a 3 cycle L1 cache which can do 2 reads at a 8
times bigger cache.
Imagine it's a 32 bits executable. In 64 bits difference will be another
10% more in advantage AMD.
I will be amazed if i see P4 beowulf clusters in future getting constructed.
This difference between dual core P4 and dual core opteron is just too big.
It's so much faster for opteron, it's no longer a compare. It's a jet
aircraft versus a propellor aircraft competing for speed.
When does intel release a follow up for P4, with some decent L1 cache size
and 2 simultaneous reads from L1?
Even an on die memory controller isn't so important like a bigger and
faster L1. 4 cycle L1 is just TOO slow for prescott/dual core P4/new Xeons.
The old P4EE's were 2 cycle L1. At least that was a good speed for L1 cache.
A 4 cycle L1 requires at least 2 or 3 simultaneously reads and 256KB L1
cache for serious software to have any impact.
More information about the Beowulf