Re: Re: ppc vs. i386
This is a nice test for the sort of thing your doing. At their web site they have a lot of published results so you can just reference them for doing the comparisons after running the suite on your machine.
More info and the database can be found at http://scl.ameslab.gov/Projects/HINT
_ _
| | _ _ _ _____ TM
|-- | | |\ | | | |
| --| | | \ | |
| | | | \| |
^ ^ ^ ^ ^ ^
*** The HINT PERFORMANCE ANALYZER ***
Version 1.0 June 1994
John L. Gustafson & Quinn O. Snell
Scalable Computing Laboratory
236 Wilhelm, Iowa State University
Ames, Iowa 50011-3020
(515) 294 - 9294
Copyright (C) 1994 Iowa State University Research Foundation, Inc.
Please send results and questions to: hint@scl.ameslab.gov
When sending results please follow the form in README
_
Double_______________________________________________________
RECT is 72 bytes
Apparent number of bits of accuracy: 53
Maximum associative whole number: 9007199254740991
Maximum number of bits of index: 30
Maximum representable index: 1073741824
Index-limited data accuracy: 53 bits
Maximum usable whole number: 9007199254740991
Grid: 67108864 wide by 134217728 high.
...........................................................
Done with first pass. Now computing net QUIPS
Finished with 9718261.166300 net QUIPs
_ _
Float________________________________________________________
RECT is 36 bytes
Apparent number of bits of accuracy: 24
Maximum associative whole number: 16777215
Maximum number of bits of index: 30
Maximum representable index: 1073741824
Index-limited data accuracy: 24 bits
Maximum usable whole number: 16777215
Grid: 4096 wide by 4096 high.
Precision is not sufficient for > 1.0 second runs.
...................................
Done with first pass. Now computing net QUIPS
Finished with 7717569.264323 net QUIPs
_ _
Integer________________________________________________________
RECT is 36 bytes
Apparent number of bits of accuracy: 31
Maximum associative whole number: 2147483647
Maximum number of bits of index: 30
Maximum representable index: 1073741824
Index-limited data accuracy: 31 bits
Maximum usable whole number: 2147483647
Grid: 32768 wide by 65536 high.
Precision is not sufficient for > 1.0 second runs.
............................................
Done with first pass. Now computing net QUIPS
Finished with 13464666.294098 net QUIPs
_________________________________________________________
This is the processor information (early model PowerBook G3)
processor : 0
cpu : 750
temperature : 0 C
clock : 231MHz
revision : 2.2
bogomips : 465.31
zero pages : total 0 (0Kb) current: 0 (0Kb) hits: 0/136 (0%)
machine : PowerBook
motherboard : AAPL,PowerBook1998 MacRISC
L2 cache : 512K unified pipelined-syncro-burst
memory : 96MB
pmac-generation : OldWorld
On Monday, October 23, 2000, at 09:22 AM, Michael Schmitz wrote:
> > > ive found my G3/400 to be about on par with my celery 466, the only
> > > thing that seems slower on ppc to me is the compiler, and the
> > > framebuffer (compared to VGA textmode which is not fair ;-))
> >
> > That's about what I see (scaling the performance to equal CPU clock).
>
> Where are you guys getting this? Just based on observation? Just for
> the record, my upgraded G3/350 appears (visually) much faster than
> my PII/366. Both machines use XFree86 4.0, kernel 2.4+, and the
> woody packages.
I don't care about how a system 'feels', or about X performance. What I
run on the machines in the lab (and on my Lombard) is a scientific
software package for restrained molecular dynamics. The RAM requirements
are modest (with under a thousand atoms) and disk I/O isn't that much
either.
I've got a few other apps that are more integer bound, I'll need to get
some test data to get measurable run times before I can post any hard
figures.
My ranking based on the lab machines, setting the G3 to a relative
performance of 1.0, and scaling all others to the same CPU clock speed:
Pentium II/III 1.14
Celeron 1.07
G3 1.0
AMD K6 0.68-0.74 (three identical K6 in identical boards)
Cyrix MII 0.60
The first two are from dual processor SMP boxes so it's possible that
single processor ones would be a bit slower. This proves nothing (other
than you can find both faster and slower Intel processors to compare
against).
Michael
--
To UNSUBSCRIBE, email to debian-powerpc-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Reply to: