Apple Power Mac G5 User Manual

Page 30

Advertising
background image

HMMer
HMMer is another application that demonstrates the performance advantages of the
Power Mac G5 with Velocity Engine for processor-intensive scientific analysis. HMMer
is a genome sequence–matching application that uses Hidden Markov Models (HMMs)
to identify similarities in genetic structures—a critical task in areas such as speech
recognition and protein and DNA analysis. By representing the properties of a
sequence family as a statistic, an HMM makes it possible to perform highly sensitive
database searches.

Erik Lindahl of Stanford University has optimized the standard HMMer source code
for the Velocity Engine. The core routines of HMMer repeatedly perform the same
operation on large amounts of data. Utilizing single-instruction, multiple-data (SIMD)
technology, the Velocity Engine enables the application to perform the same operation
on four pieces of data in a single clock cycle. With Lindahl’s optimized code, the per-
formance of a HMMer search is now 1.5 to 2.4 times faster than with a 3.2GHz Pentium
4–based PC.

To test the performance of the HMMer code, Apple searched for an HMM created from
a 358-residue sequence in the protein databank (PDB) and measured the time to
search the entire PDB.

The dual 2GHz and dual 1.8GHz Power Mac G5 performed the HMMer search 136% and

134% faster, respectively, than the 3.2GHz Pentium 4–based system, clearly demonstrating

the advantages of the Velocity Engine and symmetric multiprocessing.

1

The Dell Dimension XPS, Alienware Aurora, and Dell Precision 650 ran HMMer on Red Hat Linux.

Power Mac G5

Dual 2GHz PowerPC G5

Power Mac G5

Dual 1.8GHz PowerPC G5

Dell Dimension XPS

3.2GHz Pentium 4

Power Mac G5

1.6GHz PowerPC G5

53% faster

Dell Precision 650

Dual 3.2GHz Xeon

83% faster

Alienware Aurora

2.2GHz AMD Athlon 64 FX-51

19% slower

Genome sequence matching

Percent faster than Pentium 4

Baseline

136% faster

134% faster

30

Technology and
Performance Overview
Power Mac G5

HMMer 2.3.2 Results

Advertising