Date: 11 May 1996 16:22:41 GMT From: hafner@suncog2.forwiss.tu-muenchen.de (Walter 'madhouse' Hafner) To: freebsd-questions@freefall.FreeBSD.org Subject: Re: Highperformance X server machine Q Message-ID: <HAFNER.96May11182242@suncog2.forwiss.tu-muenchen.de> In-Reply-To: kuku@gilberto.physik.RWTH-Aachen.DE's message of 11 May 1996 17:26:52 %2B0200 References: <199605111510.RAA18673@gilberto.physik.rwth-aachen.de>
next in thread | previous in thread | raw e-mail | index | archive | help
>>On 11 May 1996 17:26:52 +0200, kuku@gilberto.physik.RWTH-Aachen.DE (Christoph P. Kukulies) said:
CPK> I just had a talk with a guy who wanted to buy a used 24bit plane
CPK> HP 715/50 and I advised him to take FreeBSD and a well equipped
CPK> P166 with a good graphics card. He was a bit reluctant when he
CPK> first thought I wanted to recommend Linux and XFree86 :-).
CPK> Any recommendations ? He wants to do image processing on that machine.
Well, we DO nothing but image processing on our machines ... Among other
types we got HP's, Sparcs and a P100 with a Matrox Millenium.
Our institute developed a high performance image processing tool (HORUS)
with about 700 image processing operators. Our "standard" benchmark is a
series of filter operations on an image. The main thing the benchmark
does is a "laws" filter on a byte image. The result is a long
image. Then it does "laws" on the long image with a byte image as
result, and so on. So the benchmark tests mainly the integer and the
memory throughput of the processor (which are BY FAR the most important
operations in image processing. Floating point is rarely used ...)
I let the benchmark results speak for themselves:
HP 712/60: 17.6s
HP 735/100: 11.7s
Sparc/10: 17.9s
Sparc/20: 12.1s
Ultra Sparc (32 bit): 4.9s
Ultra Sparc (64 bit): 4.5s
SGI Indigo2: 6.4s
SGI Indy: 15.9s
DEC 3000 Modell 600 AXP (175 MHz): 10.0s
DEC AlphaStation 600 5/266: 5.4s
Pentium 100MHz (FreeBSD 2.0.5): 12.3s
Pentium 100MHz (Windows NT): 19.2s
We get roughly the same results, when we run our most complex
application (detection and tracking of humans in image sequences) on the
machines: On the SS20 and the P100 we need about 1.6 s/image. On the
Ultra and the Alpha we need about 0.7 s/image.
Oh, yes: The Pentium was in 24 bit mode under XAccel 1.2, while all
other machines were in 8 bit mode.
So I recommend against the 715 (depending on teh price, of course). He
should buy a Pentium/Pentium-Pro with a fast graphic card and XAccel.
Or an Ultra, of course. :-)
My 2 cents,
-Walter
PS: Note the difference between FreeBSD and NT on the same
machine. *GRIN*
--
Walter Hafner____________________________ hafner@forwiss.tu-muenchen.de
FORWISS Muenchen, FG Kognitive Systeme, Raum O-134, Tel: 089/48095-220
Netscape*documentColorsHavePriority: false
Netscape*blinkingEnabled: false
<A href=http://www.forwiss.tu-muenchen.de/~hafner/>*CLICK*</A>;
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?HAFNER.96May11182242>