After upgrading to gcs-2.95 last snapshot (6/29/99) and compiling
glibc-2.1.1 with it I did some preliminary tests mainly with bzip2 that
showed me that the C compiler world under linux is really improving. So I
decided to do some "real benchmarking". I compiled almost everything
(XFree86, XEmacs, ncurses, KDE, GNOME, Gimp, Perl) with -O3
-fomit-frame-pointer -march=pentiumpro even the kernel (2.2.10-ac9 + devfs)
compiles without problem, after that I tried the BYTEmark Native Mode
Benchmark, nbench, to see the "real" differences. And I must admit that I was
very surprised.
The results speak for themselves:
Dual PII 450 Mhz (although the benchmark is not multithreaded).
Supermicro P6DGU
512 Mb RAM
linux-2.2.10-ac9 + devfs (SMP)
Using the original nbech binary that comes with nbench-byte-2.1 the results were:
TEST : Iterations/sec. : Old Index : New Index
: : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT : 197.79 : 5.07 : 1.67
STRING SORT : 18.046 : 8.06 : 1.25
BITFIELD : 5.8876e+07 : 10.10 : 2.11
FP EMULATION : 11.887 : 5.70 : 1.32
FOURIER : 3624.7 : 4.12 : 2.32
ASSIGNMENT : 2.5974 : 9.88 : 2.56
IDEA : 414.01 : 6.33 : 1.88
HUFFMAN : 198.1 : 5.49 : 1.75
NEURAL NET : 3.8976 : 6.26 : 2.63
LU DECOMPOSITION : 118.73 : 6.15 : 4.44
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX : 6.983
FLOATING-POINT INDEX: 5.415
Baseline (MSDOS*) : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
==============================LINUX DATA BELOW===============================
C compiler : gcc version 2.7.2.3
libc : libc.so.5.4.38
MEMORY INDEX : 1.890
INTEGER INDEX : 1.640
FLOATING-POINT INDEX: 3.003
Baseline (LINUX) : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38
Using the compiled version (-O3 -fomit-frame-pointer -funroll-loops) with gcc-2.95
and glibc-2.1.1 (also compiled with gcc-2.95) the results were:
TEST : Iterations/sec. : Old Index : New Index
: : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT : 278.4 : 7.14 : 2.34
STRING SORT : 19.18 : 8.57 : 1.33
BITFIELD : 7.307e+07 : 12.53 : 2.62
FP EMULATION : 13.2 : 6.33 : 1.46
FOURIER : 5151.3 : 5.86 : 3.29
ASSIGNMENT : 3.1825 : 12.11 : 3.14
IDEA : 581.4 : 8.89 : 2.64
HUFFMAN : 235.02 : 6.52 : 2.08
NEURAL NET : 9054.4 : 14545.16 : 6118.22
LU DECOMPOSITION : 194.32 : 10.07 : 7.27
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX : 8.575
FLOATING-POINT INDEX: 95.013
Baseline (MSDOS*) : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
==============================LINUX DATA BELOW===============================
C compiler : gcc version gcc-2.95
libc : unknown version
MEMORY INDEX : 2.218
INTEGER INDEX : 2.083
FLOATING-POINT INDEX: 52.698
Baseline (LINUX) : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38
quite impressive ... although the NEURAL NET result maybe wrong... I don't belive
in an almost 3000x speed up.
- german
<>-------------------------------------+-----------------------------------<>
One O.S. to rule them all, | German Gomez Garcia
One O.S. to find them. | german@pinon.ccu.uniovi.es
One O.S. to bring them all |
and in the darkness bind them. | "Wur Qanar Wur Stilor Wur Kas"
<>-------------------------------------+-----------------------------------<>
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/