Subject: Re: 6x86-166+ with NetBSD
To: William O Ferry <WOFerry+@CMU.EDU>
From: Curt Sampson <cjs@portal.ca>
List: port-i386
Date: 04/17/1997 13:41:22
On Wed, 16 Apr 1997, William O Ferry wrote:

>     I'm currently running NetBSD on an ASUS SP3G (486DX4/100)
> motherboard.

I ran on that for a couple of years before I upgraded, and was
pretty happy with it.

> I'm looking at getting the ASUS P55T2P4 motherboard, with the SC-200 NCR
> card and either an intel P133 or a Cyrix 6x86-166+.

This is exactly what I upgraded to, with the P166+. :-) I'm pretty
happy with it, too.

I also have another machine with the same motherboard and amount
of cache (512K), but an Intel P133 CPU inside. Here are the results
of a benchmark suite. I've calculated the percentage that the P166+
was better than the P133. A true P166 would theoretically be about
125% of the speed of a a P133.

c4: 107%
    3871322 pos / 22.2 sec = 174.1 Kpos/sec
    3871322 pos / 23.9 sec = 162.2 Kpos/sec

clinpack: 68%
    Rolled Double  Precision 11061 Kflops ; 100 Reps 
    Unrolled Double  Precision 10718 Kflops ; 100 Reps 
    Rolled Single  Precision 11962 Kflops ; 100 Reps 
    Unrolled Single  Precision 12103 Kflops ; 100 Reps 

    Rolled Double  Precision 16057 Kflops ; 100 Reps 
    Unrolled Double  Precision 15379 Kflops ; 100 Reps 
    Rolled Single  Precision 17210 Kflops ; 100 Reps 
    Unrolled Single  Precision 17888 Kflops ; 100 Reps 

dhry21: 114%
    Microseconds for one run through Dhrystone:     4.9 
    Dhrystones per Second:                        205902.7 
    VAX MIPS rating =    117.190 

    Microseconds for one run through Dhrystone:     5.6 
    Dhrystones per Second:                        179222.5 
    VAX MIPS rating =    102.005 

fft: 85%
    Run Time (sec) =     1.68506
    Run Time (sec) =     1.42970

flops: 67%
    MFLOPS(4)       =    17.8059
    MFLOPS(4)       =    26.4338

hanoi: 102%
    Average Moves Per 25 usec =    71.5617
    Average Moves Per 25 usec =    70.2015

heapsort: 95% / 72%
    High MIPS =    59.56    Low  MIPS =    38.43
    High MIPS =    62.45    Low  MIPS =    52.92

queens: 103%
    Run Time (sec) =    34.054
    Run Time (sec) =    35.241

nsieve: 118% / 57%
    Average RunTime =    0.038 (sec)
    High  MIPS      =    100.3
    Low   MIPS      =     24.9

    Average RunTime =    0.032 (sec)
    High  MIPS      =     85.1
    Low   MIPS      =     43.7

sim: 113%
    Run Time (sec) =     395.5
    Run Time (sec) =     446.7

tfftdp: 72%
    BenchTime (sec) =    14.4454
    VAX_FFTs =      9.737

    BenchTime (sec) =    10.3947
    VAX_FFTs =     13.532

You'll note that the P166+ is really only marginally better than
a P133 at Integer arithmetic, and considerably worse at floating
point. There also may be some cache-related things happening here,
since on benchmarks with large datasets, the larger the dataset
gets, the worse, comparatively, the P166+ performs. I should run
lmbench at some point to find out.

The P166+ is also marginally cheaper than a P133, but only by about
5% around here. I don't know if I'd buy another one.

cjs

Curt Sampson    cjs@portal.ca	   Info at http://www.portal.ca/
Internet Portal Services, Inc.	   Through infinite myst, software reverberates
Vancouver, BC  (604) 257-9400	   In code possess'd of invisible folly.