>>>>> "Yan-Song" == Yan-Song Chen <YChen@UH.EDU> writes:
Yan-Song> Hi, I would like to know some data about floating point
Yan-Song> performance of alpha based linux boxes. How fast does it
Yan-Song> compare to the pentinum based linux boxes? -- yansong
Yan-Song> chen -- YChen@uh.edu
Hm, I don't know about these other high-end board like Cabriolet
and EB64, but I could tell you something about the AXPpci33 board. I
use this with an 233 MHz processor and the fp performance isn't fixed
to one single value. I had used clinpack for comparison. I had used it
on a Pentium 100 with burst cache and 16MB and my system with 1MB
Cache and 32MB.
On the Pentium we reached aprox. 13 MFLOPS(Mega Floating-point Per
Second) which beats several workstations here on our institute. My
AXPpci33 at home was quite different. Sometimes I evaluated valued in
the range from 7-14,15 MFLOPS. The average lies around 13-14.
In my humble opinion this wouldn't made it worth to by an Alpha, but
you had to keep in mind, that Linux on an Intel is quite optimized and
sometimes it's dirty hacking. I believe if all is optimized for Alpha
or at least better as now, you will catch a much faster system as an
Intel. Considering the last big developments in the last few month I
believe that in a few month even the cheap and slow Noname (aka
AXPpci33) will run circles around the Pentium. I have no real
comparison for the so called Pentium Pro, but it will have take a
while till it catch the 21164 300 MHz. You could retrieve the sources
for clinpack from our machine:
ftp://satchmo.physik.tu-berlin.de/pub/benchmark/linpack
I hope this will help you in some way for your decision.
--
Torsten Zirzlaff
Technical University Berlin
Department of Theoretical Physics
phone: +49-30-314-26442
fax: +49-30-314-21130 e-mail :
s-mail: Hardenbergstr.36 (office) zirzlaff@satchmo.physik.tu-berlin.de
Office No. PN716 (work) ZIRZLAFF@pc.inpro.de
Sekr. PN 7-1 (private) zirzlaff@nienor.in-berlin.de
10623 Berlin, Germany
(official) http://satchmo.physik.tu-berlin.de/~zirzlaff
/-----------------------------------------------------------------------------\
| physicists do it with charm |
\-----------------------------------------------------------------------------/