Phoebe benchmarks
cl...@acorn.com
information has fallen into our hands and we thought we would share it with
you.....
Key:
Phoebe is 64 MHz IOMD2 ASIC, 228 MHz SA-110 rev T
Ursula is RISC OS 4
RPC is Risc PC with 228 MHz SA-110
RO3.7 is RISC OS 3.71
1) Dhrystones
RPC/RO3.7 = 385 kDhry/s
Phoebe/Ursula = 427 kDhrys/s = 1.1 x faster
Dhrystones is regarded as easily cacheable, and so expected to be no faster
until faster SA is fitted to Phoebe. In practice, there is a bottleneck with
writes to memory through the write buffer, and this is why Phoebe/Ursula is
faster.
2) Delaunay Triangulation
RPC/RO3.7 = 2127 Del pts/s
Phoebe/Ursula = 2695 Del pts/s = 1.3 x faster
Delaunay triangulation is a task from, eg, graphics problems. It is a
combination of list processing and significant floating point work. It is
reasonably cacheable. The speed increase is from RAM speed and Ursula 32-bit
clean FPE.
3) Wimp Nulls
RPC/RO3.7 = 180 nulls/s
Phoebe/Ursula = 4356 nulls/s = 24 x faster
Wimp Nulls is a measure of OS task swapping speed, designed to model the
OS running applications with very large Wimp slots. The test runs two
!Nulls, each with a 28 Mb Wimp slot, and the measurement is the total number
of task swaps per second achieved by the kernel. The speed up is from Phoebe
fast cache flush area, and from Ursula Lazy task swapping. The comparison
is very charitable to RPC, since the wimp slots were actually only 15 Mb on
that platform (lack of memory).
4) OS loading
RPC/RO3.7 = 80.0 s elapsed time
Phoebe/Ursula = 0.9 s elapsed time = 88.9 x faster
The OS loading test is designed to model a heavily loaded machine doing
tasks. The test does:
- create 750 dynamic areas
- create 7500 system variables
- filer_open a directory containing 75 new applications
The speed up is from Phoebe RAM and from Ursula kernel and FileSwitch
optimisations. Because ADFS is missing a few days work, the Phoebe/Ursula
score is hampered by running the hard disc inefficiently (slow PIO mode).
-- This information, whilst it is believed to be true, is not directly from
Acorn Computers Ltd and is supplied for your interest only.
-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/rg_mkgrp.xp Create Your Own Free Member Forum
Greg Hennessy
[SNIP]
Cue more discussion on what might have been :-)
greg
Andy McMullon
And what may still be.......
:-)
--
Andy: skyp...@bigfoot.com / http://www.mcfamily.demon.co.uk
Ian Hawkins
>> [SNIP]
>>
>> Cue more discussion on what might have been :-)
>
>And what may still be.......
>
>:-)
>
[drools at non-fp benchmark results]
I want one!
Regards,
Ian Hawkins
Ian.H...@william-baird.com
did you miss it?
Ian.H...@william-baird.com
Spam away! :)
Kell Gatherer
cmk...@cix.compulink.co.uk (Greg Hennessy) wrote:
>
>On Wed, 23 Sep 1998 09:15:04 GMT, cl...@acorn.com wrote:
>
>
>Cue more discussion on what might have been :-)
Aargh! Where do I find this article?
--
Kell Gatherer
ke...@locationworks.com
www.locationworks.com
dgs
> cmk...@cix.compulink.co.uk (Greg Hennessy) wrote:
...
> >On Wed, 23 Sep 1998 09:15:04 GMT, cl...@acorn.com wrote:
...
> >Cue more discussion on what might have been :-)
>
> Aargh! Where do I find this article?
You can ask me to re-post it, or you can ask me to e-mail it
to you, or you can look in Dejanews (it wasn't hidden).
It's still sitting on my hard disk, either way.
Manchester Acorn User Group - http://www.acorn.manchester.ac.uk/
RPC x86 Card Info Pages - http://acorn.cybervillage.co.uk/pccard/
"Your machine is NOT dead until it stops working" - Ian Gledhill
Stuart Bell
> 2) Delaunay Triangulation
>
> RPC/RO3.7 = 2127 Del pts/s
>
> Phoebe/Ursula = 2695 Del pts/s = 1.3 x faster
>
> Delaunay triangulation is a task from, eg, graphics problems. It is a
> combination of list processing and significant floating point work. It is
> reasonably cacheable. The speed increase is from RAM speed and Ursula 32-bit
> clean FPE.
There has been surprisingly little comment about these benchmarks.
The above seems to be the one which might be closest to giving a raw
performance figure for Phoebe, although I recognise that it's far from
perfect as a benchmark.
Are we supposed to be impressed, amazed, disheartened, or what?
cheers,
--
Stuart Bell
writing from a Wintel-free zone.
Nad
post the original message?
x-http-user-agent: mozilla/4.01 (compatible; Acorn Phoenix 2.06 (09-Jul-98);
Risc OS 3.70) Acorn-HTTP/0.82
Hmmmmmm
Nad
--
--
n...@mindless.com
http://www.argonet.co.uk/users/nad
Senior Technician - Calderdale Council Local Education Authority
Greg Hennessy <cmk...@cix.compulink.co.uk> wrote in message
360ac8ae...@nntp.netcomuk.co.uk...
>On Wed, 23 Sep 1998 09:15:04 GMT, cl...@acorn.com wrote:
>
>[SNIP]
>
>Cue more discussion on what might have been :-)
>
>
>greg
>
Paul Corke
<URL:mailto:n...@mindless.com> wrote:
> x-http-user-agent: mozilla/4.01 (compatible; Acorn Phoenix 2.06 (09-Jul-98);
> Risc OS 3.70) Acorn-HTTP/0.82
And? Phoenix is just the internal Acorn version fo Browse.
<sings> Gareth uses Browse... nah nah...
Paul.
--
mailto:pa...@ims-bristol.co.uk http://www.ims-cdc.demon.co.uk/
Charlie Baylis
Stuart Bell <sab...@argonet.co.uk> wrote:
> There has been surprisingly little comment about these benchmarks.
There doesn't seem to be much point in commenting on it really, as it
makes little difference how fast it was.
> The above seems to be the one which might be closest to giving a raw
> performance figure for Phoebe, although I recognise that it's far from
> perfect as a benchmark.
I wonder how well it does doing real things - Artworks redraw, time to
load applications, speed of the Java speccy emulator etc. MPEG3 encoding
would be good to indicate how much the new FPE helps.
If anyone takes on the Phoebe, please upclock the processor, or at least
make it possible to cut tracks on the card to upclock it!
c
--
The address in the headers is a temporary email address. To mail me use
charlie.baylis @ altavista.net (remove spaces)
Paul Firth
> >> [SNIP]
> >>
> >> Cue more discussion on what might have been :-)
> >
> >And what may still be.......
> >
> >:-)
> >
>
> [drools at non-fp benchmark results]
>
> I want one!
That dhrystone rate puts the phoebe at just under the speedof a PPRO 200MHz at
integer maths. Whats the fuss all about?
--
---------------------------------
Cheers, Paul (pa...@minds-eye.net)
---------------------------------
Will there be T.P ?
Greg Hennessy
wrote:
>That dhrystone rate puts the phoebe at just under the speedof a PPRO 200MHz at
>integer maths. Whats the fuss all about?
Makes one wonder does it not ;-). Specint figures would have been
interesting.
greg
Peter Howkins
> In article <6uc4ha$s3p$1...@plug.news.pipex.net>, Nad
> <URL:mailto:n...@mindless.com> wrote:
> > x-http-user-agent: mozilla/4.01 (compatible; Acorn Phoenix 2.06 (09-Jul-98);
> > Risc OS 3.70) Acorn-HTTP/0.82
>
> And? Phoenix is just the internal Acorn version fo Browse.
>
After the last article posted from cl...@acorn.com the same
conclusion was drawn, but now everyone knows what it is
and someone could easily fudge it. I have a program that thinks
its netscape 5. So with this being common knowedge you can't
guarantee the're from acorn.
Peter
Paul Corke
<URL:mailto:p.j.h...@lboro.ac.uk> wrote:
> its netscape 5. So with this being common knowedge you can't
> guarantee the're from acorn.
Well they did a damn good job of pursuading hotmail that they were
posting from "gtupper.acorn.com".
Rob Hemmings
the raw processor performance. Why? Surely the relatively small
improvement here is to be expected since a processor of the same speed as a
Risc PC is bound to give similar results for benchmarks that measure just
the speed of the processor. (We don't know if this was the final intended
processor anyway.)
However, the other benchmarks clearly benefit from the real
hardware/software improvements to Phoebe (e,g, bus speed) and sound to be
far more like the results that would be experienced in actually using the
machine for more general things. These figures have got to be stunning.
Now I really want a Phoebe!
Surely we can't let something that can run one of the benchmarks 88x faster
than a StrongARM Risc PC disappear?
--
-------------------------------------------------------------
Rob Hemmings Southport
Tel: +44 (0)1704 573210 ro...@argonet.co.uk
Paul Vigay
> post the original message?
>
> Risc OS 3.70) Acorn-HTTP/0.82
I thought that was just the internal code name for !Browse, so it looks like
they were using the same version that the rest of us have. Where is 2.07?
--
Paul Vigay Computer Resources Manager,
__\\|//__ Bohunt Community School
http://www.matrix.clara.net (` o-o ') Liphook, Hampshire
---------------------------------ooO-(_)-Ooo-----------------------------
All views my own and I reserve the right to change them without warning!
Remove ".vogonpoetry" to reply by email.
Thomas Boroske
Rob Hemmings <rhem...@argonet.co.uk> wrote:
> Now I really want a Phoebe!
>
> Surely we can't let something that can run one of the benchmarks 88x faster
> than a StrongARM Risc PC disappear?
Well - maybe you just want RiscOS 4. I mean, dhrystone wasn't much
faster (as expected), that other benchmark was 1.3 times faster ...
Wimp polls are massively faster (nice, of course), but like the last
test this is probably mostly down to OS improvements ?
Kind regards,
--
Thomas Boroske
Mike Stephens
Yes, Dhrystone pretty much depends on core speed as expected. Delaunay is an
example of a reasonably cache friendly task (real task, not artificial
benchmark code), and so 1.3x faster was very encouraging, at same core
speed.
The wimp nulls and OS loading both benefit significantly from the hardware,
but Ursula combines with this to give big-time numbers. There would be no
improvement of wimp nulls from the Ursula changes on all (most?) Risc PC's
because rev S or earlier StrongARMs have a bug that defeats the Lazy scheme.
Phoebe was to be fitted with rev T StrongARMs initially.
Some real world and much less cache friendly tasks were tried on the next
morning after these results were released. They went something like (the
figures are probably a bit rough, but not wildly out):
Compilation
RPC/3.7 = 11 mins elapsed time
Phoebe/Ursula = 5 mins elapsed time = 2.2 x faster
Compilation is compile and link to library (in a task window) of detailed
model of SA-110, written in C. The large times are mainly consumed in
compiling two of the larger/more complex C files. The speed is partly from
Ursula Lazy task swapping, but mostly from Phoebe hardware.
Emulation
RPC/3.7 = 0.6 kHz
Phoebe/Ursula = 1.5 kHz = 2.5 x faster
Emulation is using SA-110 model, inside a machine model, running an OS ROM.
Because of its massive complexity/size, the SA-110 model is (ironically)
very unfriendly to the SA-110 primary caches. There is virtually no Ursula
impact on this speed; it is almost entirely from Phoebe SDRAM.
Regards,
Mike.
Theo Markettos
p.j.h...@lboro.ac.uk (Peter Howkins) wrote:
> Paul Corke wrote:
>
> > In article <6uc4ha$s3p$1...@plug.news.pipex.net>, Nad
> > <URL:mailto:n...@mindless.com> wrote:
> > > x-http-user-agent: mozilla/4.01 (compatible; Acorn Phoenix 2.06 (09-Jul-98);
> > > Risc OS 3.70) Acorn-HTTP/0.82
> >
> > And? Phoenix is just the internal Acorn version fo Browse.
> >
>
> After the last article posted from cl...@acorn.com the same
> conclusion was drawn, but now everyone knows what it is
> and someone could easily fudge it. I have a program that thinks
> its netscape 5. So with this being common knowedge you can't
> guarantee the're from acorn.
However, it also had:
X-Http-Proxy: 1.0 x7.dejanews.com:80 (Squid/1.1.22) for client 136.170.128.80
A while ago someone from Acorn filled in a form on my Web site - Demon's
script gave me this:
This data is mailed from your web page
The host that sent this request was rmanby.acorn.co.uk [136.170.128.75]
(although if a proxy was used, you will get the address of the proxy instead)
The user's browser reported itself to be
Mozilla/4.01 (Compatible; Acorn Phoenix 2.05 [21-Apr-98];
RISC OS 3.70) Acorn-HTTP/0.79
Note how the IP addresses are rather similar. I would agree that otherwise
such an article could easily be faked, but according to Dejanews' tracking
information, this seems to be genuine - at least the poster does - that
doesn't guarantee the information in the post to be accurate, however...
--
Theo Markettos theoma...@letterbox.com
Liphook
Hampshire Web site, including Acorn backup software
UK http://www.marketto.demon.co.uk/
The Sherratt Clan
Theo Markettos <th...@marketto.demon.co.uk> wrote:
> In article <360A78D7...@lboro.ac.uk>,
> However, it also had:
> X-Http-Proxy: 1.0 x7.dejanews.com:80 (Squid/1.1.22) for client
> 136.170.128.80
host reports that 136.170.128.80 is gtupper.acorn.co.uk
Any use?
TTFN, Karl
Rainer M Duffner
If you've got a CD to share with us - don't hesitate...
Do you have specINT somewhere sitting on the shelf ?
From what I remember, some of the "benchmarks" would be "fully featured"
programs under RISC-OS.
Only too slow, I'm afraid. :-(
cheers,
Rainer
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|Rainer Duffner, E-Mail: duf...@fh-konstanz.de |
| & Rainer....@konstanz.netsurf.de |
|Fachhochschule Konstanz, Germany |
|"What's a Network ?" - Bill Gates, early 1980s |
| WWW:http://www-stud.fh-konstanz.de/~duffner |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Rainer M Duffner
<URL:mailto:kshe...@argonet.co.uk> wrote:
> host reports that 136.170.128.80 is gtupper.acorn.co.uk
>
> Any use?
What do they say on the anonymizer-site ?
"Because on today's internet, people know you're a dog"
:-)
I'd use an oversea's shell-account to do things like that....
Thomas Boroske
Mike Stephens <mi...@sulis.demon.co.uk> wrote:
> Compilation
>
> RPC/3.7 = 11 mins elapsed time
> Phoebe/Ursula = 5 mins elapsed time = 2.2 x faster
>
> Compilation is compile and link to library (in a task window) of detailed
> model of SA-110, written in C. The large times are mainly consumed in
> compiling two of the larger/more complex C files. The speed is partly from
> Ursula Lazy task swapping, but mostly from Phoebe hardware.
>
> Emulation
>
> RPC/3.7 = 0.6 kHz
> Phoebe/Ursula = 1.5 kHz = 2.5 x faster
>
> Emulation is using SA-110 model, inside a machine model, running an OS ROM.
> Because of its massive complexity/size, the SA-110 model is (ironically)
> very unfriendly to the SA-110 primary caches. There is virtually no Ursula
> impact on this speed; it is almost entirely from Phoebe SDRAM.
Thanks for the info, looks like what we expected from Phoebe.
Andy McMullon
Exactly!
I don't know why people were getting excited - either slagging off the
poor perfermance increase or going exstatic about certain allegedly
amazing increases.
We expected that Phoebe/Ursula would take out the bottlenecks caused by
the bus and iron out some of the major creases in the OS.
We always expected that significant increases in processor based tasks
would wait on faster (or more) StrongArms.
We'll have to wait patiently to see what PB and the consortium pull out
of this particular hat now!
The Sherratt Clan
Rainer M Duffner <Rainer....@konstanz.netsurf.de> wrote:
> In article <488bc62bf...@argonet.co.uk>, The Sherratt Clan
> <URL:mailto:kshe...@argonet.co.uk> wrote:
> > host reports that 136.170.128.80 is gtupper.acorn.co.uk
> >
> > Any use?
> What do they say on the anonymizer-site ?
> "Because on today's internet, people know you're a dog"
> :-)
Actually, I'm a fishŚ
> I'd use an oversea's shell-account to do things like that....
If it is/was anonymised I suspect that the owner/user of
gtupper.acorn.co.uk is a bit cheesed offŚ
TTFN, Karl.
Greg Hennessy
<Rainer....@konstanz.netsurf.de> wrote:
esting.
>
>If you've got a CD to share with us - don't hesitate...
>Do you have specINT somewhere sitting on the shelf ?
>From what I remember, some of the "benchmarks" would be "fully featured"
>programs under RISC-OS.
>Only too slow, I'm afraid. :-(
Spec is a bit more complicated :-) that that. See
for further details.
greg
Stuart Halliday
> 4) OS loading
>
> RPC/RO3.7 = 80.0 s elapsed time
>
> Phoebe/Ursula = 0.9 s elapsed time = 88.9 x faster
Wow. That is fast!
So the desktop operation just got very much faster. Incredible.
--
Stuart Halliday
Acorn Cybervillage
http://acorn.cybervillage.co.uk/
Stuart Halliday
> > post the original message?
> >
> > Risc OS 3.70) Acorn-HTTP/0.82
>
> they were using the same version that the rest of us have. Where is 2.07?
I thought the Browser RISCOS development got cancelled back in May?